chore: remove non-weight files (batch 5)

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/template_single_spk/voc1/local/data_prep.sh

@@ -1,87 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-# shellcheck disable=SC1091
-. ./path.sh || exit 1;
-
-fs=22050
-num_dev=100
-num_eval=100
-train_set="train_nodev"
-dev_set="dev"
-eval_set="eval"
-shuffle=false
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-db_root=$1
-data_dir=$2
-
-# check arguments
-if [ $# != 2 ]; then
-    echo "Usage: $0 [Options] <db_root> <data_dir>"
-    echo ""
-    echo "Options:"
-    echo "    --num_dev: number of development uttreances (default=250)."
-    echo "    --num_eval: number of evaluation uttreances (default=250)."
-    echo "    --train_set: name of train set (default=train_nodev)."
-    echo "    --dev_set: name of dev set (default=dev)."
-    echo "    --eval_set: name of eval set (default=eval)."
-    echo "    --shuffle: whether to perform shuffle in making dev / eval set (default=false)."
-    exit 1
-fi
-
-set -euo pipefail
-
-[ ! -e "${data_dir}/all" ] && mkdir -p "${data_dir}/all"
-
-# set filenames
-scp="${data_dir}/all/wav.scp"
-
-# check file existence
-[ -e "${scp}" ] && rm "${scp}"
-
-# make all scp
-find "${db_root}" -follow -name "*.wav" | sort | while read -r filename; do
-    id=$(basename "${filename}" | sed -e "s/\.[^\.]*$//g")
-    echo "${id} cat ${filename} | sox -t wav - -c 1 -b 16 -t wav - rate ${fs} |" >> "${scp}"
-done
-
-# split
-num_all=$(wc -l < "${scp}")
-num_deveval=$((num_dev + num_eval))
-num_train=$((num_all - num_deveval))
-if [ ${num_eval} -ne 0 ]; then
-    utils/split_data.sh \
-        --num_first "${num_train}" \
-        --num_second "${num_deveval}" \
-        --shuffle "${shuffle}" \
-        "${data_dir}/all" \
-        "${data_dir}/${train_set}" \
-        "${data_dir}/deveval"
-    utils/split_data.sh \
-        --num_first "${num_dev}" \
-        --num_second "${num_eval}" \
-        --shuffle "${shuffle}" \
-        "${data_dir}/deveval" \
-        "${data_dir}/${dev_set}" \
-        "${data_dir}/${eval_set}"
-else
-    utils/split_data.sh \
-        --num_first "${num_train}" \
-        --num_second "${num_deveval}" \
-        --shuffle "${shuffle}" \
-        "${data_dir}/all" \
-        "${data_dir}/${train_set}" \
-        "${data_dir}/${dev_set}"
-    cp -r "${data_dir}/${dev_set}" "${data_dir}/${eval_set}"
-fi
-
-# remove tmp directories
-rm -rf "${data_dir}/all"
-rm -rf "${data_dir}/deveval"
-
-echo "Successfully prepared data."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/template_single_spk/voc1/path.sh

@@ -1,33 +0,0 @@
-# cuda related
-export CUDA_HOME=/usr/local/cuda-11.3
-export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
-
-# path related
-export PRJ_ROOT="${PWD}/../../.."
-if [ -e "${PRJ_ROOT}/tools/venv/bin/activate" ]; then
-    # shellcheck disable=SC1090
-    . "${PRJ_ROOT}/tools/venv/bin/activate"
-fi
-
-# python related
-export OMP_NUM_THREADS=1
-export PYTHONIOENCODING=UTF-8
-export MPL_BACKEND=Agg
-
-# check installation
-if ! command -v parallel-wavegan-train > /dev/null; then
-    echo "Error: It seems setup is not finished." >&2
-    echo "Error: Please setup your environment by following README.md" >&2
-    return 1
-fi
-if ! command -v jq > /dev/null; then
-    echo "Error: It seems jq is not installed." >&2
-    echo "Error: Please install via \`sudo apt-get install jq\`." >&2
-    echo "Error: If you do not have sudo, please download from https://stedolan.github.io/jq/download/." >&2
-    return 1
-fi
-if ! command -v yq > /dev/null; then
-    echo "Error: It seems yq is not installed." >&2
-    echo "Error: Please install via \`pip install yq\`." >&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/template_single_spk/voc1/run.sh

@@ -1,188 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-. ./cmd.sh || exit 1;
-. ./path.sh || exit 1;
-
-# basic settings
-stage=0        # stage to start
-stop_stage=100 # stage to stop
-verbose=1      # verbosity level (lower is less info)
-n_gpus=1       # number of gpus in training
-n_jobs=4       # number of parallel jobs in feature extraction
-
-# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
-conf=conf/parallel_wavegan.v1.yaml
-
-# directory path setting
-db_root=/path/to/database # direcotry including wavfiles (MODIFY BY YOURSELF)
-                          # each wav filename in the directory should be unique
-                          # e.g.
-                          # /path/to/database
-                          # ├── utt_1.wav
-                          # ├── utt_2.wav
-                          # │   ...
-                          # └── utt_N.wav
-dumpdir=dump # directory to dump features
-
-# subset setting
-shuffle=false # whether to shuffle the data to create subset
-num_dev=100   # the number of development data
-num_eval=100  # the number of evaluation data
-              # (if set to 0, the same dev set is used as eval set)
-
-# training related setting
-tag=""     # tag for directory to save model
-resume=""  # checkpoint path to resume training
-           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
-pretrain="" # checkpoint path to load pretrained parameters
-            # (e.g. ../../jsut/<path>/<to>/checkpoint-400000steps.pkl)
-
-# decoding related setting
-checkpoint="" # checkpoint path to be used for decoding
-              # if not provided, the latest one will be used
-              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-train_set="train_nodev" # name of training data directory
-dev_set="dev"           # name of development data direcotry
-eval_set="eval"         # name of evaluation data direcotry
-
-set -euo pipefail
-
-if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
-    echo "Stage 0: Data preparation"
-    local/data_prep.sh \
-        --fs "$(yq ".sampling_rate" "${conf}")" \
-        --num_dev "${num_dev}" \
-        --num_eval "${num_eval}" \
-        --train_set "${train_set}" \
-        --dev_set "${dev_set}" \
-        --eval_set "${eval_set}" \
-        --shuffle "${shuffle}" \
-        "${db_root}" data
-fi
-
-stats_ext=$(grep -q "hdf5" <(yq ".format" "${conf}") && echo "h5" || echo "npy")
-if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
-    echo "Stage 1: Feature extraction"
-    # extract raw features
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
-        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
-        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
-            parallel-wavegan-preprocess \
-                --config "${conf}" \
-                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
-                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished feature extraction of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished feature extraction."
-
-    # calculate statistics for normalization
-    if [ -z "${pretrain}" ]; then
-        # calculate statistics for normalization
-        echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics.log."
-        ${train_cmd} "${dumpdir}/${train_set}/compute_statistics.log" \
-            parallel-wavegan-compute-statistics \
-                --config "${conf}" \
-                --rootdir "${dumpdir}/${train_set}/raw" \
-                --dumpdir "${dumpdir}/${train_set}" \
-                --verbose "${verbose}"
-        echo "Successfully finished calculation of statistics."
-    else
-        echo "Use statistics of pretrained model. Skip statistics computation."
-        cp "$(dirname "${pretrain}")/stats.${stats_ext}" "${dumpdir}/${train_set}"
-    fi
-
-    # normalize and dump them
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/norm" ] && mkdir -p "${dumpdir}/${name}/norm"
-        echo "Nomalization start. See the progress via ${dumpdir}/${name}/norm/normalize.*.log."
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/norm/normalize.JOB.log" \
-            parallel-wavegan-normalize \
-                --config "${conf}" \
-                --stats "${dumpdir}/${train_set}/stats.${stats_ext}" \
-                --rootdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --dumpdir "${dumpdir}/${name}/norm/dump.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished normalization of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished normalization."
-fi
-
-if [ -z "${tag}" ]; then
-    expdir="exp/${train_set}_$(basename "${conf}" .yaml)"
-    if [ -n "${pretrain}" ]; then
-        pretrain_tag=$(basename "$(dirname "${pretrain}")")
-        expdir+="_${pretrain_tag}"
-    fi
-else
-    expdir="exp/${train_set}_${tag}"
-fi
-if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
-    echo "Stage 2: Network training"
-    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
-    cp "${dumpdir}/${train_set}/stats.${stats_ext}" "${expdir}"
-    if [ "${n_gpus}" -gt 1 ]; then
-        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
-    else
-        train="parallel-wavegan-train"
-    fi
-    echo "Training start. See the progress via ${expdir}/train.log."
-    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
-        ${train} \
-            --config "${conf}" \
-            --train-dumpdir "${dumpdir}/${train_set}/norm" \
-            --dev-dumpdir "${dumpdir}/${dev_set}/norm" \
-            --outdir "${expdir}" \
-            --resume "${resume}" \
-            --pretrain "${pretrain}" \
-            --verbose "${verbose}"
-    echo "Successfully finished training."
-fi
-
-if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
-    echo "Stage 3: Network decoding"
-    # shellcheck disable=SC2012
-    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
-        [ "${n_gpus}" -gt 1 ] && n_gpus=1
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
-        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
-            parallel-wavegan-decode \
-                --dumpdir "${dumpdir}/${name}/norm" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished decoding."
-fi
-echo "Finished."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/cmd.sh

@@ -1,91 +0,0 @@
-# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
-# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
-# e.g.
-#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
-#
-# Options:
-#   --time <time>: Limit the maximum time to execute.
-#   --mem <mem>: Limit the maximum memory usage.
-#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
-#   --num-threads <ngpu>: Specify the number of CPU core.
-#   --gpu <ngpu>: Specify the number of GPU devices.
-#   --config: Change the configuration file from default.
-#
-# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
-# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
-# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
-# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
-#
-# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
-# These options are mapping to specific options for each backend and
-# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
-# If jobs failed, your configuration might be wrong for your environment.
-#
-#
-# The official documentaion for run.pl, queue.pl, slurm.pl, and ssh.pl:
-#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
-# =========================================================~
-
-
-# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
-cmd_backend="local"
-
-# Local machine, without any Job scheduling system
-if [ "${cmd_backend}" = local ]; then
-
-    # The other usage
-    export train_cmd="utils/run.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/run.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/run.pl"
-
-# Local machine, without any Job scheduling system
-elif [ "${cmd_backend}" = stdout ]; then
-
-    # The other usage
-    export train_cmd="utils/stdout.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/stdout.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/stdout.pl"
-
-# "qsub" (SGE, Torque, PBS, etc.)
-elif [ "${cmd_backend}" = sge ]; then
-    # The default setting is written in conf/queue.conf.
-    # You must change "-q g.q" for the "queue" for your environment.
-    # To know the "queue" names, type "qhost -q"
-    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
-
-    export train_cmd="utils/queue.pl"
-    export cuda_cmd="utils/queue.pl"
-    export decode_cmd="utils/queue.pl"
-
-# "sbatch" (Slurm)
-elif [ "${cmd_backend}" = slurm ]; then
-    # The default setting is written in conf/slurm.conf.
-    # You must change "-p cpu" and "-p gpu" for the "partion" for your environment.
-    # To know the "partion" names, type "sinfo".
-    # You can use "--gpu * " by defualt for slurm and it is interpreted as "--gres gpu:*"
-    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
-
-    export train_cmd="utils/slurm.pl"
-    export cuda_cmd="utils/slurm.pl"
-    export decode_cmd="utils/slurm.pl"
-
-elif [ "${cmd_backend}" = ssh ]; then
-    # You have to create ".queue/machines" to specify the host to execute jobs.
-    # e.g. .queue/machines
-    #   host1
-    #   host2
-    #   host3
-    # Assuming you can login them without any password, i.e. You have to set ssh keys.
-
-    export train_cmd="utils/ssh.pl"
-    export cuda_cmd="utils/ssh.pl"
-    export decode_cmd="utils/ssh.pl"
-
-else
-    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/conf/hifigan_hubert.v1.yaml

@@ -1,176 +0,0 @@
-# This configuration is based on HiFiGAN V1, derived
-# from official repository (https://github.com/jik876/hifi-gan).
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 16000     # Sampling rate.
-fft_size: null           # FFT size.
-hop_size: 320            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: null             # Window function.
-num_mels: 2              # Number of mel basis.
-fmin: null               # Minimum freq in mel basis calculation.
-fmax: null               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: DiscreteSymbolHiFiGANGenerator
-generator_params:
-    in_channels: 512                      # Number of input channels.
-    out_channels: 1                       # Number of output channels.
-    channels: 512                         # Number of initial channels.
-    num_embs: 100
-    num_spk_embs: 128
-    spk_emb_dim: 512
-    concat_spk_emb: false
-    kernel_size: 7                        # Kernel size of initial and final conv layers.
-    upsample_scales: [10, 8, 2, 2]        # Upsampling scales.
-    upsample_kernel_sizes: [20, 16, 4, 4] # Kernel size for upsampling layers.
-    resblock_kernel_sizes: [3, 7, 11]     # Kernel size for residual blocks.
-    resblock_dilations:                   # Dilations for residual blocks.
-        - [1, 3, 5]
-        - [1, 3, 5]
-        - [1, 3, 5]
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
-    nonlinear_activation: "LeakyReLU"     # Nonlinear activation type.
-    nonlinear_activation_params:          # Nonlinear activation paramters.
-        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
-discriminator_params:
-    scales: 3                              # Number of multi-scale discriminator.
-    scale_downsample_pooling: "AvgPool1d"  # Pooling operation for scale discriminator.
-    scale_downsample_pooling_params:
-        kernel_size: 4                     # Pooling kernel size.
-        stride: 2                          # Pooling stride.
-        padding: 2                         # Padding size.
-    scale_discriminator_params:
-        in_channels: 1                     # Number of input channels.
-        out_channels: 1                    # Number of output channels.
-        kernel_sizes: [15, 41, 5, 3]       # List of kernel sizes.
-        channels: 128                      # Initial number of channels.
-        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        max_groups: 16                     # Maximum number of groups in downsampling conv layers.
-        bias: true
-        downsample_scales: [4, 4, 4, 4, 1] # Downsampling scales.
-        nonlinear_activation: "LeakyReLU"  # Nonlinear activation.
-        nonlinear_activation_params:
-            negative_slope: 0.1
-    follow_official_norm: true             # Whether to follow the official norm setting.
-    periods: [2, 3, 5, 7, 11]              # List of period for multi-period discriminator.
-    period_discriminator_params:
-        in_channels: 1                     # Number of input channels.
-        out_channels: 1                    # Number of output channels.
-        kernel_sizes: [5, 3]               # List of kernel sizes.
-        channels: 32                       # Initial number of channels.
-        downsample_scales: [3, 3, 3, 3, 1] # Downsampling scales.
-        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        bias: true                         # Whether to use bias parameter in conv layer."
-        nonlinear_activation: "LeakyReLU"  # Nonlinear activation.
-        nonlinear_activation_params:       # Nonlinear activation paramters.
-            negative_slope: 0.1
-        use_weight_norm: true              # Whether to apply weight normalization.
-        use_spectral_norm: false           # Whether to apply spectral normalization.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-use_stft_loss: false # Whether to use multi-resolution STFT loss.
-use_mel_loss: true   # Whether to use Mel-spectrogram loss.
-mel_loss_params:     # Mel-spectrogram loss parameters.
-    fs: 16000
-    fft_size: 1024
-    hop_size: 256
-    win_length: null
-    window: "hann"
-    num_mels: 80
-    fmin: 0
-    fmax: 8000
-    log_base: null   # Log base. If set to null, use natural logarithm.
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-use_feat_match_loss: true
-feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-    average_by_layers: false         # Whether to average loss by #layers in each discriminator.
-    include_final_outputs: true      # Whether to include final outputs in feat match loss calculation.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_aux: 45.0       # Loss balancing coefficient for STFT loss.
-lambda_adv: 1.0        # Loss balancing coefficient for adversarial loss.
-lambda_feat_match: 2.0 # Loss balancing coefficient for feat match loss..
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 16              # Batch size.
-batch_max_steps: 10240      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: Adam
-generator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-generator_scheduler_type: MultiStepLR
-generator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-generator_grad_norm: -1
-discriminator_optimizer_type: Adam
-discriminator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-discriminator_scheduler_type: MultiStepLR
-discriminator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-discriminator_grad_norm: -1
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-generator_train_start_steps: 1     # Number of steps to start to train discriminator.
-discriminator_train_start_steps: 0 # Number of steps to start to train discriminator.
-train_max_steps: 2500000           # Number of training steps.
-save_interval_steps: 50000         # Interval steps to save checkpoint.
-eval_interval_steps: 1000          # Interval steps to evaluate the network.
-log_interval_steps: 100            # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/conf/hifigan_hubert_24k.v1.yaml

@@ -1,176 +0,0 @@
-# This configuration is based on HiFiGAN V1, derived
-# from official repository (https://github.com/jik876/hifi-gan).
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000     # Sampling rate.
-fft_size: null           # FFT size.
-hop_size: 480            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: null             # Window function.
-num_mels: 2              # Number of mel basis.
-fmin: null               # Minimum freq in mel basis calculation.
-fmax: null               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: DiscreteSymbolHiFiGANGenerator
-generator_params:
-    in_channels: 512                      # Number of input channels.
-    out_channels: 1                       # Number of output channels.
-    channels: 512                         # Number of initial channels.
-    num_embs: 100
-    num_spk_embs: 128
-    spk_emb_dim: 512
-    concat_spk_emb: false
-    kernel_size: 7                        # Kernel size of initial and final conv layers.
-    upsample_scales: [12, 10, 2, 2]        # Upsampling scales.
-    upsample_kernel_sizes: [24, 20, 4, 4] # Kernel size for upsampling layers.
-    resblock_kernel_sizes: [3, 7, 11]     # Kernel size for residual blocks.
-    resblock_dilations:                   # Dilations for residual blocks.
-        - [1, 3, 5]
-        - [1, 3, 5]
-        - [1, 3, 5]
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
-    nonlinear_activation: "LeakyReLU"     # Nonlinear activation type.
-    nonlinear_activation_params:          # Nonlinear activation paramters.
-        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
-discriminator_params:
-    scales: 3                              # Number of multi-scale discriminator.
-    scale_downsample_pooling: "AvgPool1d"  # Pooling operation for scale discriminator.
-    scale_downsample_pooling_params:
-        kernel_size: 4                     # Pooling kernel size.
-        stride: 2                          # Pooling stride.
-        padding: 2                         # Padding size.
-    scale_discriminator_params:
-        in_channels: 1                     # Number of input channels.
-        out_channels: 1                    # Number of output channels.
-        kernel_sizes: [15, 41, 5, 3]       # List of kernel sizes.
-        channels: 128                      # Initial number of channels.
-        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        max_groups: 16                     # Maximum number of groups in downsampling conv layers.
-        bias: true
-        downsample_scales: [4, 4, 4, 4, 1] # Downsampling scales.
-        nonlinear_activation: "LeakyReLU"  # Nonlinear activation.
-        nonlinear_activation_params:
-            negative_slope: 0.1
-    follow_official_norm: true             # Whether to follow the official norm setting.
-    periods: [2, 3, 5, 7, 11]              # List of period for multi-period discriminator.
-    period_discriminator_params:
-        in_channels: 1                     # Number of input channels.
-        out_channels: 1                    # Number of output channels.
-        kernel_sizes: [5, 3]               # List of kernel sizes.
-        channels: 32                       # Initial number of channels.
-        downsample_scales: [3, 3, 3, 3, 1] # Downsampling scales.
-        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        bias: true                         # Whether to use bias parameter in conv layer."
-        nonlinear_activation: "LeakyReLU"  # Nonlinear activation.
-        nonlinear_activation_params:       # Nonlinear activation paramters.
-            negative_slope: 0.1
-        use_weight_norm: true              # Whether to apply weight normalization.
-        use_spectral_norm: false           # Whether to apply spectral normalization.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-use_stft_loss: false # Whether to use multi-resolution STFT loss.
-use_mel_loss: true   # Whether to use Mel-spectrogram loss.
-mel_loss_params:     # Mel-spectrogram loss parameters.
-    fs: 24000
-    fft_size: 2048
-    hop_size: 300
-    win_length: 1200
-    window: "hann"
-    num_mels: 80
-    fmin: 80
-    fmax: 7600
-    log_base: null   # Log base. If set to null, use natural logarithm.
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-use_feat_match_loss: true
-feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-    average_by_layers: false         # Whether to average loss by #layers in each discriminator.
-    include_final_outputs: true      # Whether to include final outputs in feat match loss calculation.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_aux: 45.0       # Loss balancing coefficient for STFT loss.
-lambda_adv: 1.0        # Loss balancing coefficient for adversarial loss.
-lambda_feat_match: 2.0 # Loss balancing coefficient for feat match loss..
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 16              # Batch size.
-batch_max_steps: 12000      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: Adam
-generator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-generator_scheduler_type: MultiStepLR
-generator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-generator_grad_norm: -1
-discriminator_optimizer_type: Adam
-discriminator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-discriminator_scheduler_type: MultiStepLR
-discriminator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-discriminator_grad_norm: -1
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-generator_train_start_steps: 1     # Number of steps to start to train discriminator.
-discriminator_train_start_steps: 0 # Number of steps to start to train discriminator.
-train_max_steps: 2500000           # Number of training steps.
-save_interval_steps: 20000         # Interval steps to save checkpoint.
-eval_interval_steps: 1000          # Interval steps to evaluate the network.
-log_interval_steps: 100            # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/conf/slurm.conf

@@ -1,12 +0,0 @@
-# Default configuration
-command sbatch --export=PATH  --ntasks-per-node=1
-option time=* --time $0
-option mem=* --mem-per-cpu $0
-option mem=0          # Do not add anything to qsub_opts
-option num_threads=* --cpus-per-task $0 --ntasks-per-node=1
-option num_threads=1 --cpus-per-task 1  --ntasks-per-node=1 # Do not add anything to qsub_opts
-default gpu=0
-option gpu=0 -p cpu
-option gpu=* -p gpu --gres=gpu:$0
-# note: the --max-jobs-run option is supported as a special case
-# by slurm.pl and you don't have to handle it in the config file.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/conf/style_melgan_hubert.v1.yaml

@@ -1,145 +0,0 @@
-# This configuration is based on StyleMelGAN paper.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 16000     # Sampling rate.
-fft_size: null           # FFT size.
-hop_size: 320            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: null             # Window function.
-num_mels: 1              # Number of mel basis.
-fmin: null               # Minimum freq in mel basis calculation.
-fmax: null               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. " npy " or " hdf5 " is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "DiscreteSymbolStyleMelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 128
-    aux_channels: 128
-    channels: 64
-    out_channels: 1
-    num_embs: 100
-    num_spk_embs: 128
-    spk_emb_dim: 128
-    concat_spk_emb: false
-    kernel_size: 9
-    dilation: 2
-    bias: True
-    noise_upsample_scales: [7, 2, 2, 2]
-    noise_upsample_activation: "LeakyReLU"
-    noise_upsample_activation_params:
-        negative_slope: 0.2
-    upsample_scales: [5, 2, 2, 2, 2, 2, 2, 1, 1]
-    upsample_mode: "nearest"
-    gated_function: "softmax"
-    use_weight_norm: True
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "StyleMelGANDiscriminator" # Discriminator type.
-discriminator_params:
-    repeats: 4
-    window_sizes: [512, 1024, 2048, 4096]
-    pqmf_params:
-        - [1, None, None, None]
-        - [2, 62, 0.26700, 9.0]
-        - [4, 62, 0.14200, 9.0]
-        - [8, 62, 0.07949, 9.0]
-    discriminator_params:
-        out_channels: 1
-        kernel_sizes: [5, 3]
-        channels: 16
-        max_downsample_channels: 512
-        bias: True
-        downsample_scales: [4, 4, 4, 1]
-        nonlinear_activation: "LeakyReLU"
-        nonlinear_activation_params:
-            negative_slope: 0.2
-    use_weight_norm: True
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-lambda_aux: 1.0                   # Loss balancing coefficient for aux loss.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 1.0 # Loss balancing coefficient for adv loss.
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 16              # Batch size.
-batch_max_steps: 17920      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: Adam
-generator_optimizer_params:
-    lr: 1.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-generator_scheduler_type: MultiStepLR
-generator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 100000
-        - 300000
-        - 500000
-        - 700000
-        - 900000
-generator_grad_norm: -1
-discriminator_optimizer_type: Adam
-discriminator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-discriminator_scheduler_type: MultiStepLR
-discriminator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-discriminator_grad_norm: -1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
-train_max_steps: 1500000                # Number of training steps.
-save_interval_steps: 50000              # Interval steps to save checkpoint.
-eval_interval_steps: 1000              # Interval steps to evaluate the network.
-log_interval_steps: 100                # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/path.sh

@@ -1,33 +0,0 @@
-# cuda related
-export CUDA_HOME=/usr/local/cuda-11.3
-export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
-
-# path related
-export PRJ_ROOT="${PWD}/../../.."
-if [ -e "${PRJ_ROOT}/tools/venv/bin/activate" ]; then
-    # shellcheck disable=SC1090
-    . "${PRJ_ROOT}/tools/venv/bin/activate"
-fi
-
-# python related
-export OMP_NUM_THREADS=1
-export PYTHONIOENCODING=UTF-8
-export MPL_BACKEND=Agg
-
-# check installation
-if ! command -v parallel-wavegan-train > /dev/null; then
-    echo "Error: It seems setup is not finished." >&2
-    echo "Error: Please setup your environment by following README.md" >&2
-    return 1
-fi
-if ! command -v jq > /dev/null; then
-    echo "Error: It seems jq is not installed." >&2
-    echo "Error: Please install via \`sudo apt-get install jq\`." >&2
-    echo "Error: If you do not have sudo, please download from https://stedolan.github.io/jq/download/." >&2
-    return 1
-fi
-if ! command -v yq > /dev/null; then
-    echo "Error: It seems yq is not installed." >&2
-    echo "Error: Please install via \`pip install yq\`." >&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/hubert_voc1/run.sh

@@ -1,151 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-. ./cmd.sh || exit 1;
-. ./path.sh || exit 1;
-
-# basic settings
-stage=-1       # stage to start
-stop_stage=100 # stage to stop
-verbose=1      # verbosity level (lower is less info)
-n_gpus=1       # number of gpus in training
-n_jobs=16      # number of parallel jobs in feature extraction
-
-# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
-conf=conf/parallel_wavegan_hubert.v1.yaml
-
-# directory path setting
-dumpdir=dump           # directory to dump features
-
-# training related setting
-tag=""     # tag for directory to save model
-resume=""  # checkpoint path to resume training
-           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
-
-# decoding related setting
-checkpoint="" # checkpoint path to be used for decoding
-              # if not provided, the latest one will be used
-              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)
-
-train_set=tr_2mix_trim
-dev_set=cv_2mix
-eval_set=tt_2mix
-
-hubert_text=downloads/dump_feature_hubert9/label_km100_vctk_16k_tr_2mix_trim/all.txt
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-set -euo pipefail
-
-if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
-    echo "Stage 0: Data preparation"
-    echo "Data preparation is not prepared."
-    echo "Please manually locate kaldi-style directories as follows:"
-    cat << EOF
-$ ls data/${train_set}
-wav.scp  utt2spk
-$ ls data/${dev_set}
-wav.scp  utt2spk
-$ ls data/${eval_set}
-wav.scp  utt2spk
-EOF
-    # shellcheck disable=SC2016
-    echo 'Please make sure "${hubert_text}" is correctly set.'
-    echo "Then, please run from stage 1 via --stage 1."
-    exit 1;
-fi
-
-if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
-    echo "Stage 1: Feature extraction"
-    if [ ! -e "${hubert_text}" ]; then
-        echo "Valid --hubert_text is not provided. Please prepare it by yourself."
-        echo "hubert_text should be like kaldi-style text as follows:"
-        cat << EOF
-utt_id_1 0 0 0 0 1 1 1 1 2 2 2 2
-utt_id_2 0 0 0 0 0 0 3 3 3 3 3 3 5 5 5 5
-...
-EOF
-        exit 1
-    fi
-    # extract raw features
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
-        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
-        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
-            local/preprocess_hubert.py \
-                --config "${conf}" \
-                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
-                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --utt2spk "data/${name}/utt2spk" \
-                --spk2idx "data/${train_set}/spk2idx" \
-                --text "${hubert_text}" \
-                --verbose "${verbose}"
-        echo "Successfully finished feature extraction of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished feature extraction."
-
-fi
-
-if [ -z "${tag}" ]; then
-    expdir="exp/${train_set}_vctk_$(basename "${conf}" .yaml)"
-else
-    expdir="exp/${train_set}_vctk_${tag}"
-fi
-if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
-    echo "Stage 2: Network training"
-    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
-    if [ "${n_gpus}" -gt 1 ]; then
-        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
-    else
-        train="parallel-wavegan-train"
-    fi
-    # shellcheck disable=SC2012
-    resume="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    echo "Training start. See the progress via ${expdir}/train.log."
-    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
-        ${train} \
-            --config "${conf}" \
-            --train-dumpdir "${dumpdir}/${train_set}/raw" \
-            --dev-dumpdir "${dumpdir}/${dev_set}/raw" \
-            --outdir "${expdir}" \
-            --resume "${resume}" \
-            --verbose "${verbose}"
-    echo "Successfully finished training."
-fi
-
-if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
-    echo "Stage 3: Network decoding"
-    # shellcheck disable=SC2012
-    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
-        [ "${n_gpus}" -gt 1 ] && n_gpus=1
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
-        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
-            parallel-wavegan-decode \
-                --dumpdir "${dumpdir}/${name}/raw" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished decoding."
-fi
-echo "Finished."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/cmd.sh

@@ -1,91 +0,0 @@
-# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
-# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
-# e.g.
-#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
-#
-# Options:
-#   --time <time>: Limit the maximum time to execute.
-#   --mem <mem>: Limit the maximum memory usage.
-#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
-#   --num-threads <ngpu>: Specify the number of CPU core.
-#   --gpu <ngpu>: Specify the number of GPU devices.
-#   --config: Change the configuration file from default.
-#
-# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
-# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
-# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
-# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
-#
-# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
-# These options are mapping to specific options for each backend and
-# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
-# If jobs failed, your configuration might be wrong for your environment.
-#
-#
-# The official documentaion for run.pl, queue.pl, slurm.pl, and ssh.pl:
-#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
-# =========================================================~
-
-
-# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
-cmd_backend="local"
-
-# Local machine, without any Job scheduling system
-if [ "${cmd_backend}" = local ]; then
-
-    # The other usage
-    export train_cmd="utils/run.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/run.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/run.pl"
-
-# Local machine, without any Job scheduling system
-elif [ "${cmd_backend}" = stdout ]; then
-
-    # The other usage
-    export train_cmd="utils/stdout.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/stdout.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/stdout.pl"
-
-# "qsub" (SGE, Torque, PBS, etc.)
-elif [ "${cmd_backend}" = sge ]; then
-    # The default setting is written in conf/queue.conf.
-    # You must change "-q g.q" for the "queue" for your environment.
-    # To know the "queue" names, type "qhost -q"
-    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
-
-    export train_cmd="utils/queue.pl"
-    export cuda_cmd="utils/queue.pl"
-    export decode_cmd="utils/queue.pl"
-
-# "sbatch" (Slurm)
-elif [ "${cmd_backend}" = slurm ]; then
-    # The default setting is written in conf/slurm.conf.
-    # You must change "-p cpu" and "-p gpu" for the "partion" for your environment.
-    # To know the "partion" names, type "sinfo".
-    # You can use "--gpu * " by defualt for slurm and it is interpreted as "--gres gpu:*"
-    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
-
-    export train_cmd="utils/slurm.pl"
-    export cuda_cmd="utils/slurm.pl"
-    export decode_cmd="utils/slurm.pl"
-
-elif [ "${cmd_backend}" = ssh ]; then
-    # You have to create ".queue/machines" to specify the host to execute jobs.
-    # e.g. .queue/machines
-    #   host1
-    #   host2
-    #   host3
-    # Assuming you can login them without any password, i.e. You have to set ssh keys.
-
-    export train_cmd="utils/ssh.pl"
-    export cuda_cmd="utils/ssh.pl"
-    export decode_cmd="utils/ssh.pl"
-
-else
-    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/conf/hifigan.v1.yaml

@@ -1,180 +0,0 @@
-# This is the configuration file for VCTK dataset.
-# This configuration is based on HiFiGAN V1, which is
-# an official configuration. But I found that the optimizer
-# setting does not work well with my implementation.
-# So I changed optimizer settings as follows:
-# - AdamW -> Adam
-# - betas: [0.8, 0.99] -> betas: [0.5, 0.9]
-# - Scheduler: ExponentialLR -> MultiStepLR
-# To match the shift size difference, the upsample scales
-# is also modified from the original 256 shift setting.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000     # Sampling rate.
-fft_size: 2048           # FFT size.
-hop_size: 300            # Hop size.
-win_length: 1200         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 7600               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: HiFiGANGenerator
-generator_params:
-    in_channels: 80                       # Number of input channels.
-    out_channels: 1                       # Number of output channels.
-    channels: 512                         # Number of initial channels.
-    kernel_size: 7                        # Kernel size of initial and final conv layers.
-    upsample_scales: [5, 5, 4, 3]         # Upsampling scales.
-    upsample_kernel_sizes: [10, 10, 8, 6] # Kernel size for upsampling layers.
-    resblock_kernel_sizes: [3, 7, 11]     # Kernel size for residual blocks.
-    resblock_dilations:                   # Dilations for residual blocks.
-        - [1, 3, 5]
-        - [1, 3, 5]
-        - [1, 3, 5]
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
-    nonlinear_activation: "LeakyReLU"     # Nonlinear activation type.
-    nonlinear_activation_params:          # Nonlinear activation paramters.
-        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
-discriminator_params:
-    scales: 3                              # Number of multi-scale discriminator.
-    scale_downsample_pooling: "AvgPool1d"  # Pooling operation for scale discriminator.
-    scale_downsample_pooling_params:
-        kernel_size: 4                     # Pooling kernel size.
-        stride: 2                          # Pooling stride.
-        padding: 2                         # Padding size.
-    scale_discriminator_params:
-        in_channels: 1                     # Number of input channels.
-        out_channels: 1                    # Number of output channels.
-        kernel_sizes: [15, 41, 5, 3]       # List of kernel sizes.
-        channels: 128                      # Initial number of channels.
-        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        max_groups: 16                     # Maximum number of groups in downsampling conv layers.
-        bias: true
-        downsample_scales: [4, 4, 4, 4, 1] # Downsampling scales.
-        nonlinear_activation: "LeakyReLU"  # Nonlinear activation.
-        nonlinear_activation_params:
-            negative_slope: 0.1
-    follow_official_norm: true             # Whether to follow the official norm setting.
-    periods: [2, 3, 5, 7, 11]              # List of period for multi-period discriminator.
-    period_discriminator_params:
-        in_channels: 1                     # Number of input channels.
-        out_channels: 1                    # Number of output channels.
-        kernel_sizes: [5, 3]               # List of kernel sizes.
-        channels: 32                       # Initial number of channels.
-        downsample_scales: [3, 3, 3, 3, 1] # Downsampling scales.
-        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        bias: true                         # Whether to use bias parameter in conv layer."
-        nonlinear_activation: "LeakyReLU"  # Nonlinear activation.
-        nonlinear_activation_params:       # Nonlinear activation paramters.
-            negative_slope: 0.1
-        use_weight_norm: true              # Whether to apply weight normalization.
-        use_spectral_norm: false           # Whether to apply spectral normalization.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-use_stft_loss: false                 # Whether to use multi-resolution STFT loss.
-use_mel_loss: true                   # Whether to use Mel-spectrogram loss.
-mel_loss_params:
-    fs: 24000
-    fft_size: 2048
-    hop_size: 300
-    win_length: 1200
-    window: "hann"
-    num_mels: 80
-    fmin: 0
-    fmax: 12000
-    log_base: null
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-use_feat_match_loss: true
-feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-    average_by_layers: false         # Whether to average loss by #layers in each discriminator.
-    include_final_outputs: false     # Whether to include final outputs in feat match loss calculation.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_aux: 45.0       # Loss balancing coefficient for STFT loss.
-lambda_adv: 1.0        # Loss balancing coefficient for adversarial loss.
-lambda_feat_match: 2.0 # Loss balancing coefficient for feat match loss..
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 16              # Batch size.
-batch_max_steps: 8400       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: Adam
-generator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-generator_scheduler_type: MultiStepLR
-generator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-generator_grad_norm: -1
-discriminator_optimizer_type: Adam
-discriminator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-discriminator_scheduler_type: MultiStepLR
-discriminator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-discriminator_grad_norm: -1
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-generator_train_start_steps: 1     # Number of steps to start to train discriminator.
-discriminator_train_start_steps: 0 # Number of steps to start to train discriminator.
-train_max_steps: 2500000           # Number of training steps.
-save_interval_steps: 10000         # Interval steps to save checkpoint.
-eval_interval_steps: 1000          # Interval steps to evaluate the network.
-log_interval_steps: 100            # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/conf/multi_band_melgan.v2.yaml

@@ -1,154 +0,0 @@
-# This is the hyperparameter configuration file for MelGAN.
-# Please make sure this is adjusted for the VCTK dataset. If you want to
-# apply to the other dataset, you might need to carefully change some parameters.
-# This configuration requires ~ 6GB memory and will finish within 3 days on Titan V.
-
-# This configuration is based on multi-band MelGAN but the hop size and sampling
-# rate is different from the paper (16kHz vs 24kHz). The number of iteraions
-# is now shown in the paper so currently we train 1M iterations (not sure enough
-# to converge). Compared to multi_band_melgan.v1 config, Adam optimizer without
-# gradient norm is used, which is based on @dathudeptrai advice.
-# https://github.com/kan-bayashi/ParallelWaveGAN/issues/143#issuecomment-632539906
-
-# We found that the use of small batch_max_steps has no bad effect for the quality,
-# so we use 8,100 for batch_max_steps to accelerate the training.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000     # Sampling rate.
-fft_size: 2048           # FFT size.
-hop_size: 300            # Hop size.
-win_length: 1200         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 7600               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 2048    # Frame size in trimming.
-trim_hop_size: 512       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "MelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 80               # Number of input channels.
-    out_channels: 4               # Number of output channels.
-    kernel_size: 7                # Kernel size of initial and final conv layers.
-    channels: 384                 # Initial number of channels for conv layers.
-    upsample_scales: [5, 5, 3]    # List of Upsampling scales.
-    stack_kernel_size: 3          # Kernel size of dilated conv layers in residual stack.
-    stacks: 4                     # Number of stacks in a single residual stack module.
-    use_weight_norm: True         # Whether to use weight normalization.
-    use_causal_conv: False        # Whether to use causal convolution.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
-discriminator_params:
-    in_channels: 1                    # Number of input channels.
-    out_channels: 1                   # Number of output channels.
-    scales: 3                         # Number of multi-scales.
-    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
-    downsample_pooling_params:        # Parameters of the above pooling function.
-        kernel_size: 4
-        stride: 2
-        padding: 1
-        count_include_pad: False
-    kernel_sizes: [5, 3]              # List of kernel size.
-    channels: 16                      # Number of channels of the initial conv layer.
-    max_downsample_channels: 512      # Maximum number of channels of downsampling layers.
-    downsample_scales: [4, 4, 4]      # List of downsampling scales.
-    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
-    nonlinear_activation_params:      # Parameters of nonlinear activation function.
-        negative_slope: 0.2
-    use_weight_norm: True             # Whether to use weight norm.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-use_subband_stft_loss: true
-subband_stft_loss_params:
-    fft_sizes: [384, 683, 171]  # List of FFT size for STFT-based loss.
-    hop_sizes: [30, 60, 10]     # List of hop size for STFT-based loss
-    win_lengths: [150, 300, 60] # List of window length for STFT-based loss.
-    window: "hann_window"       # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-use_feat_match_loss: false # Whether to use feature matching loss.
-lambda_adv: 2.5            # Loss balancing coefficient for adversarial loss.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 64             # Batch size.
-batch_max_steps: 8100      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
-num_workers: 4             # Number of workers in Pytorch DataLoader.
-remove_short_samples: true # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: "Adam"       # Generator's optimizer type.
-generator_optimizer_params:
-    lr: 1.0e-3                          # Generator's learning rate.
-    eps: 1.0e-7                         # Generator's epsilon.
-    weight_decay: 0.0                   # Generator's weight decay coefficient.
-    amsgrad: true
-generator_grad_norm: -1                 # Generator's gradient norm.
-generator_scheduler_type: "MultiStepLR" # Generator's scheduler type.
-generator_scheduler_params:
-    gamma: 0.5                          # Generator's scheduler gamma.
-    milestones:                         # At each milestone, lr will be multiplied by gamma.
-        - 100000
-        - 200000
-        - 300000
-        - 400000
-        - 500000
-        - 600000
-discriminator_optimizer_type: "Adam"        # Discriminator's optimizer type.
-discriminator_optimizer_params:
-    lr: 1.0e-3                              # Discriminator's learning rate.
-    eps: 1.0e-7                             # Discriminator's epsilon.
-    weight_decay: 0.0                       # Discriminator's weight decay coefficient.
-    amsgrad: true
-discriminator_grad_norm: -1                 # Discriminator's gradient norm.
-discriminator_scheduler_type: "MultiStepLR" # Discriminator's scheduler type.
-discriminator_scheduler_params:
-    gamma: 0.5                              # Discriminator's scheduler gamma.
-    milestones:                             # At each milestone, lr will be multiplied by gamma.
-        - 100000
-        - 200000
-        - 300000
-        - 400000
-        - 500000
-        - 600000
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 200000 # Number of steps to start to train discriminator.
-train_max_steps: 1000000                # Number of training steps.
-save_interval_steps: 50000              # Interval steps to save checkpoint.
-eval_interval_steps: 1000               # Interval steps to evaluate the network.
-log_interval_steps: 1000                # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/conf/parallel_wavegan.v1.long.yaml

@@ -1,122 +0,0 @@
-# This is the hyperparameter configuration file for Parallel WaveGAN.
-# Please make sure this is adjusted for the VCTK corpus. If you want to
-# apply to the other dataset, you might need to carefully change some parameters.
-# This configuration requires 12 GB GPU memory and takes ~7 days on RTX TITAN.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000     # Sampling rate.
-fft_size: 2048           # FFT size.
-hop_size: 300            # Hop size.
-win_length: 1200         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 7600               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Kernel size of dilated convolution.
-    layers: 30            # Number of residual block layers.
-    stacks: 3             # Number of stacks i.e., dilation cycles.
-    residual_channels: 64 # Number of channels in residual conv.
-    gate_channels: 128    # Number of channels in gated conv.
-    skip_channels: 64     # Number of channels in skip conv.
-    aux_channels: 80      # Number of channels for auxiliary feature conv.
-                          # Must be the same as num_mels.
-    aux_context_window: 2 # Context window size for auxiliary feature.
-                          # If set to 2, previous 2 and future 2 frames will be considered.
-    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    upsample_net: "ConvInUpsampleNetwork" # Upsampling network architecture.
-    upsample_params:                      # Upsampling network parameters.
-        upsample_scales: [4, 5, 3, 5]     # Upsampling scales. Prodcut of these must be the same as hop size.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Number of output channels.
-    layers: 10            # Number of conv layers.
-    conv_channels: 64     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    nonlinear_activation: "LeakyReLU" # Nonlinear function after each conv.
-    nonlinear_activation_params:      # Nonlinear function parameters
-        negative_slope: 0.2           # Alpha in LeakyReLU.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 4.0  # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 6              # Batch size.
-batch_max_steps: 24000     # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2             # Number of workers in Pytorch DataLoader.
-remove_short_samples: true # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
-train_max_steps: 1000000                # Number of training steps.
-save_interval_steps: 5000               # Interval steps to save checkpoint.
-eval_interval_steps: 1000               # Interval steps to evaluate the network.
-log_interval_steps: 100                 # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/conf/parallel_wavegan.v1.yaml

@@ -1,122 +0,0 @@
-# This is the hyperparameter configuration file for Parallel WaveGAN.
-# Please make sure this is adjusted for the VCTK corpus. If you want to
-# apply to the other dataset, you might need to carefully change some parameters.
-# This configuration requires 12 GB GPU memory and takes ~3 days on RTX TITAN.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000     # Sampling rate.
-fft_size: 2048           # FFT size.
-hop_size: 300            # Hop size.
-win_length: 1200         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 7600               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Kernel size of dilated convolution.
-    layers: 30            # Number of residual block layers.
-    stacks: 3             # Number of stacks i.e., dilation cycles.
-    residual_channels: 64 # Number of channels in residual conv.
-    gate_channels: 128    # Number of channels in gated conv.
-    skip_channels: 64     # Number of channels in skip conv.
-    aux_channels: 80      # Number of channels for auxiliary feature conv.
-                          # Must be the same as num_mels.
-    aux_context_window: 2 # Context window size for auxiliary feature.
-                          # If set to 2, previous 2 and future 2 frames will be considered.
-    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    upsample_net: "ConvInUpsampleNetwork" # Upsampling network architecture.
-    upsample_params:                      # Upsampling network parameters.
-        upsample_scales: [4, 5, 3, 5]     # Upsampling scales. Prodcut of these must be the same as hop size.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Number of output channels.
-    layers: 10            # Number of conv layers.
-    conv_channels: 64     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    nonlinear_activation: "LeakyReLU" # Nonlinear function after each conv.
-    nonlinear_activation_params:      # Nonlinear function parameters
-        negative_slope: 0.2           # Alpha in LeakyReLU.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 4.0  # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 6              # Batch size.
-batch_max_steps: 24000     # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2             # Number of workers in Pytorch DataLoader.
-remove_short_samples: true # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
-train_max_steps: 1000000                # Number of training steps.
-save_interval_steps: 5000               # Interval steps to save checkpoint.
-eval_interval_steps: 1000               # Interval steps to evaluate the network.
-log_interval_steps: 100                 # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/conf/slurm.conf

@@ -1,12 +0,0 @@
-# Default configuration
-command sbatch --export=PATH  --ntasks-per-node=1
-option time=* --time $0
-option mem=* --mem-per-cpu $0
-option mem=0          # Do not add anything to qsub_opts
-option num_threads=* --cpus-per-task $0 --ntasks-per-node=1
-option num_threads=1 --cpus-per-task 1  --ntasks-per-node=1 # Do not add anything to qsub_opts
-default gpu=0
-option gpu=0 -p cpu
-option gpu=* -p gpu --gres=gpu:$0
-# note: the --max-jobs-run option is supported as a special case
-# by slurm.pl and you don't have to handle it in the config file.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/conf/style_melgan.v1.yaml

@@ -1,147 +0,0 @@
-# This is the configuration file for VCTK dataset.
-# This configuration is based on StyleMelGAN paper but
-# uses MSE loss instead of Hinge loss. And I found that
-# batch_size = 8 is also working good. So maybe if you
-# want to accelerate the training, you can reduce the
-# batch size (e.g. 8 or 16). Upsampling scales is modified
-# to fit the shift size 300 pt.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000     # Sampling rate.
-fft_size: 2048           # FFT size.
-hop_size: 300            # Hop size.
-win_length: 1200         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 7600               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. " npy " or " hdf5 " is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "StyleMelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 128
-    aux_channels: 80
-    channels: 64
-    out_channels: 1
-    kernel_size: 9
-    dilation: 2
-    bias: True
-    noise_upsample_scales: [10, 2, 2, 2]
-    noise_upsample_activation: "LeakyReLU"
-    noise_upsample_activation_params:
-        negative_slope: 0.2
-    upsample_scales: [5, 1, 5, 1, 3, 1, 2, 2, 1]
-    upsample_mode: "nearest"
-    gated_function: "softmax"
-    use_weight_norm: True
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "StyleMelGANDiscriminator" # Discriminator type.
-discriminator_params:
-    repeats: 4
-    window_sizes: [512, 1024, 2048, 4096]
-    pqmf_params:
-        - [1, None, None, None]
-        - [2, 62, 0.26700, 9.0]
-        - [4, 62, 0.14200, 9.0]
-        - [8, 62, 0.07949, 9.0]
-    discriminator_params:
-        out_channels: 1
-        kernel_sizes: [5, 3]
-        channels: 16
-        max_downsample_channels: 512
-        bias: True
-        downsample_scales: [4, 4, 4, 1]
-        nonlinear_activation: "LeakyReLU"
-        nonlinear_activation_params:
-            negative_slope: 0.2
-    use_weight_norm: True
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-lambda_aux: 1.0                   # Loss balancing coefficient for aux loss.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 1.0 # Loss balancing coefficient for adv loss.
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 32              # Batch size.
-batch_max_steps: 24000      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: Adam
-generator_optimizer_params:
-    lr: 1.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-generator_scheduler_type: MultiStepLR
-generator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 100000
-        - 300000
-        - 500000
-        - 700000
-        - 900000
-generator_grad_norm: -1
-discriminator_optimizer_type: Adam
-discriminator_optimizer_params:
-    lr: 2.0e-4
-    betas: [0.5, 0.9]
-    weight_decay: 0.0
-discriminator_scheduler_type: MultiStepLR
-discriminator_scheduler_params:
-    gamma: 0.5
-    milestones:
-        - 200000
-        - 400000
-        - 600000
-        - 800000
-discriminator_grad_norm: -1
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
-train_max_steps: 1500000                # Number of training steps.
-save_interval_steps: 50000              # Interval steps to save checkpoint.
-eval_interval_steps: 1000              # Interval steps to evaluate the network.
-log_interval_steps: 100                # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/local/data_download.sh

@@ -1,37 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-download_dir=$1
-
-# check arguments
-if [ $# != 1 ]; then
-    echo "Usage: $0 <download_dir>"
-    exit 1
-fi
-
-set -euo pipefail
-
-cwd=$(pwd)
-if [ ! -e "${download_dir}/VCTK-Corpus" ]; then
-    mkdir -p "${download_dir}"
-    cd "${download_dir}" || exit 1;
-    wget http://www.udialogue.org/download/VCTK-Corpus.tar.gz
-    tar xvzf ./VCTK-Corpus.tar.gz
-    rm ./VCTK-Corpus.tar.gz
-    cd "${cwd}" || exit 1;
-    echo "Successfully downloaded data."
-else
-    echo "Already exists. Skipped."
-fi
-
-if [ ! -e "${download_dir}/VCTK-Corpus/lab" ]; then
-    cd "${download_dir}" || exit 1;
-    git clone https://github.com/kan-bayashi/VCTKCorpusFullContextLabel.git
-    cp -r VCTKCorpusFullContextLabel/lab ./VCTK-Corpus
-    cd "${cwd}" || exit 1;
-    echo "Successfully downloaded label data."
-else
-    echo "Already exists. Skipped."
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/local/data_prep.sh

@@ -1,118 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-# shellcheck disable=SC1091
-. ./path.sh || exit 1;
-
-fs=24000
-num_dev=10
-num_eval=10
-train_set="train_nodev"
-dev_set="dev"
-eval_set="eval"
-shuffle=false
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-db_root=$1
-spk=$2
-data_dir=$3
-
-# check arguments
-if [ $# != 3 ]; then
-    echo "Usage: $0 [Options] <db_root> <spk> <data_dir>"
-    echo "e.g.: $0 downloads/VCTK-Corpus p225 data"
-    echo ""
-    echo "Options:"
-    echo "    --fs: target sampling rate (default=24000)."
-    echo "    --num_dev: number of development uttreances (default=10)."
-    echo "    --num_eval: number of evaluation uttreances (default=10)."
-    echo "    --train_set: name of train set (default=train_nodev)."
-    echo "    --dev_set: name of dev set (default=dev)."
-    echo "    --eval_set: name of eval set (default=eval)."
-    echo "    --shuffle: whether to perform shuffle in making dev / eval set (default=false)."
-    exit 1
-fi
-
-set -euo pipefail
-
-# check spk existence
-[ ! -e "${db_root}/lab/mono/${spk}" ] && \
-    echo "${spk} does not exist." >&2 && exit 1;
-
-[ ! -e "${data_dir}/all" ] && mkdir -p "${data_dir}/all"
-
-# set filenames
-scp="${data_dir}/all/wav.scp"
-segments="${data_dir}/all/segments"
-
-# check file existence
-[ -e "${scp}" ] && rm "${scp}"
-[ -e "${segments}" ] && rm "${segments}"
-
-# make scp and segments
-find "${db_root}/wav48/${spk}" -follow -name "*.wav" | sort | while read -r wav; do
-    id=$(basename "${wav}" | sed -e "s/\.[^\.]*$//g")
-    lab=${db_root}/lab/mono/${spk}/${id}.lab
-
-    # check lab existence
-    if [ ! -e "${lab}" ]; then
-        echo "${id} does not have a label file. skipped."
-        continue
-    fi
-
-    echo "${id} cat ${wav} | sox -t wav - -c 1 -b 16 -t wav - rate ${fs} |" >> "${scp}"
-
-    # parse start and end time from HTS-style mono label
-    idx=1
-    while true; do
-        next_idx=$((idx+1))
-        next_symbol=$(sed -n "${next_idx}p" "${lab}" | awk '{print $3}')
-        if [ "${next_symbol}" != "pau" ]; then
-            start_nsec=$(sed -n "${idx}p" "${lab}" | awk '{print $2}')
-            break
-        fi
-        idx=${next_idx}
-    done
-    idx=$(wc -l < "${lab}")
-    while true; do
-        prev_idx=$((idx-1))
-        prev_symbol=$(sed -n "${prev_idx}p" "${lab}" | awk '{print $3}')
-        if [ "${prev_symbol}" != "pau" ]; then
-            end_nsec=$(sed -n "${idx}p" "${lab}" | awk '{print $1}')
-            break
-        fi
-        idx=${prev_idx}
-    done
-    start_sec=$(echo "${start_nsec}*0.0000001" | bc | sed "s/^\./0./")
-    end_sec=$(echo "${end_nsec}*0.0000001" | bc | sed "s/^\./0./")
-    echo "${id} ${id} ${start_sec} ${end_sec}" >> "${segments}"
-done
-
-# split
-num_all=$(wc -l < "${scp}")
-num_deveval=$((num_dev + num_eval))
-num_train=$((num_all - num_deveval))
-utils/split_data.sh \
-    --num_first "${num_train}" \
-    --num_second "${num_deveval}" \
-    --shuffle "${shuffle}" \
-    "${data_dir}/all" \
-    "${data_dir}/${train_set}" \
-    "${data_dir}/deveval"
-utils/split_data.sh \
-    --num_first "${num_dev}" \
-    --num_second "${num_eval}" \
-    --shuffle "${shuffle}" \
-    "${data_dir}/deveval" \
-    "${data_dir}/${dev_set}" \
-    "${data_dir}/${eval_set}"
-
-# remove tmp directories
-rm -rf "${data_dir}/all"
-rm -rf "${data_dir}/deveval"
-
-echo "Successfully prepared data."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/local/decode_from_text.py

@@ -1,174 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-"""Decode text with trained VQ-VAE Generator or discrete symbol vocoder."""
-
-import argparse
-import logging
-import os
-import time
-
-import parallel_wavegan.models
-import soundfile as sf
-import torch
-import yaml
-from parallel_wavegan.utils import load_model
-from tqdm import tqdm
-
-
-def main():
-    """Run decoding process."""
-    parser = argparse.ArgumentParser(
-        description=(
-            "Decode text with trained VQ-VAE decoder "
-            "(See detail in parallel_wavegan/bin/decode.py)."
-        )
-    )
-    parser.add_argument(
-        "--text",
-        required=True,
-        type=str,
-        help="kaldi-style text file.",
-    )
-    parser.add_argument(
-        "--utt2spk",
-        default=None,
-        type=str,
-        help="kaldi-style utt2spk file.",
-    )
-    parser.add_argument(
-        "--spk2idx",
-        default=None,
-        type=str,
-        help="kaldi-style spk2idx file.",
-    )
-    parser.add_argument(
-        "--outdir",
-        type=str,
-        required=True,
-        help="directory to save generated speech.",
-    )
-    parser.add_argument(
-        "--checkpoint",
-        type=str,
-        required=True,
-        help="checkpoint file to be loaded.",
-    )
-    parser.add_argument(
-        "--config",
-        default=None,
-        type=str,
-        help=(
-            "yaml format configuration file. if not explicitly provided, "
-            "it will be searched in the checkpoint directory. (default=None)"
-        ),
-    )
-    parser.add_argument(
-        "--verbose",
-        type=int,
-        default=1,
-        help="logging level. higher is more logging. (default=1)",
-    )
-    args = parser.parse_args()
-
-    # set logger
-    if args.verbose > 1:
-        logging.basicConfig(
-            level=logging.DEBUG,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    elif args.verbose > 0:
-        logging.basicConfig(
-            level=logging.INFO,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    else:
-        logging.basicConfig(
-            level=logging.WARN,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-        logging.warning("Skip DEBUG/INFO messages")
-
-    # check directory existence
-    if not os.path.exists(args.outdir):
-        os.makedirs(args.outdir)
-
-    # load config
-    if args.config is None:
-        dirname = os.path.dirname(args.checkpoint)
-        args.config = os.path.join(dirname, "config.yml")
-    with open(args.config) as f:
-        config = yaml.load(f, Loader=yaml.Loader)
-    config.update(vars(args))
-
-    # setup model
-    if torch.cuda.is_available():
-        device = torch.device("cuda")
-    else:
-        device = torch.device("cpu")
-    model = load_model(args.checkpoint, config)
-    logging.info(f"Loaded model parameters from {args.checkpoint}.")
-    model.remove_weight_norm()
-    model = model.eval().to(device)
-    is_vqvae = isinstance(model, parallel_wavegan.models.VQVAE)
-
-    # setup dataset
-    with open(args.text) as f:
-        lines = [l_.replace("\n", "") for l_ in f.readlines()]
-    text = {l_.split()[0]: list(map(int, l_.split()[1:])) for l_ in lines}
-
-    utt2spk = None
-    if args.utt2spk is not None:
-        assert args.spk2idx is not None
-        with open(args.utt2spk) as f:
-            lines = [l_.replace("\n", "") for l_ in f.readlines()]
-        utt2spk = {l_.split()[0]: str(l_.split()[1]) for l_ in lines}
-        with open(args.spk2idx) as f:
-            lines = [l_.replace("\n", "") for l_ in f.readlines()]
-        spk2idx = {l_.split()[0]: int(l_.split()[1]) for l_ in lines}
-
-    # start generation
-    total_rtf = 0.0
-    with torch.no_grad(), tqdm(text.items(), desc="[decode]") as pbar:
-        for idx, items in enumerate(pbar, 1):
-            utt_id, indices = items
-            z = torch.LongTensor(indices).view(1, -1).to(device)
-            g = None
-            if utt2spk is not None:
-                spk_idx = spk2idx[utt2spk[utt_id]]
-                g = torch.tensor(spk_idx).long().view(1).to(device)
-            if is_vqvae:
-                # VQVAE case
-                start = time.time()
-                y = model.decode(z, None, g).view(-1).cpu().numpy()
-                rtf = (time.time() - start) / (len(y) / config["sampling_rate"])
-                pbar.set_postfix({"RTF": rtf})
-                total_rtf += rtf
-            else:
-                # Discrete symbol vocoder case
-                start = time.time()
-                g = int(g.item()) if g is not None else None
-                y = model.inference(z.view(-1, 1), g=g).view(-1).cpu().numpy()
-                rtf = (time.time() - start) / (len(y) / config["sampling_rate"])
-                pbar.set_postfix({"RTF": rtf})
-                total_rtf += rtf
-
-            # save as PCM 16 bit wav file
-            sf.write(
-                os.path.join(config["outdir"], f"{utt_id}_gen.wav"),
-                y,
-                config["sampling_rate"],
-                "PCM_16",
-            )
-
-    # report average RTF
-    logging.info(
-        f"Finished generation of {idx} utterances (RTF = {total_rtf / idx:.03f})."
-    )
-
-
-if __name__ == "__main__":
-    main()

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/local/preprocess_hubert.py

@@ -1,249 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-"""Perform preprocessing and raw feature extraction."""
-
-import argparse
-import logging
-import os
-
-import librosa
-import numpy as np
-import resampy
-import soundfile as sf
-import yaml
-from parallel_wavegan.datasets import AudioDataset, AudioSCPDataset
-from parallel_wavegan.utils import write_hdf5
-from tqdm import tqdm
-
-
-def main():
-    """Run preprocessing process."""
-    parser = argparse.ArgumentParser(
-        description=(
-            "Preprocess audio and then extract features (See detail in"
-            " parallel_wavegan/bin/preprocess.py)."
-        )
-    )
-    parser.add_argument(
-        "--wav-scp",
-        "--scp",
-        default=None,
-        type=str,
-        help="kaldi-style wav.scp file. you need to specify either scp or rootdir.",
-    )
-    parser.add_argument(
-        "--segments",
-        default=None,
-        type=str,
-        help=(
-            "kaldi-style segments file. if use, you must to specify both scp and"
-            " segments."
-        ),
-    )
-    parser.add_argument(
-        "--text",
-        default=None,
-        type=str,
-        help="kaldi-style text format hubert embedding index.",
-    )
-    parser.add_argument(
-        "--utt2spk",
-        default=None,
-        type=str,
-        help=(
-            "kaldi-style utt2spk file. If you want to add global conditionning with "
-            "speaker id, you need to specify this argument."
-        ),
-    )
-    parser.add_argument(
-        "--spk2idx",
-        default=None,
-        type=str,
-        help=(
-            "kaldi-style spk2idx file. If you want to add global conditionning with "
-            "speaker id, you need to specify this argument."
-        ),
-    )
-    parser.add_argument(
-        "--rootdir",
-        default=None,
-        type=str,
-        help=(
-            "directory including wav files. you need to specify either scp or rootdir."
-        ),
-    )
-    parser.add_argument(
-        "--dumpdir",
-        type=str,
-        required=True,
-        help="directory to dump feature files.",
-    )
-    parser.add_argument(
-        "--config",
-        type=str,
-        required=True,
-        help="yaml format configuration file.",
-    )
-    parser.add_argument(
-        "--verbose",
-        type=int,
-        default=1,
-        help="logging level. higher is more logging. (default=1)",
-    )
-    args = parser.parse_args()
-
-    # set logger
-    if args.verbose > 1:
-        logging.basicConfig(
-            level=logging.DEBUG,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    elif args.verbose > 0:
-        logging.basicConfig(
-            level=logging.INFO,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    else:
-        logging.basicConfig(
-            level=logging.WARN,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-        logging.warning("Skip DEBUG/INFO messages")
-
-    # load config
-    with open(args.config) as f:
-        config = yaml.load(f, Loader=yaml.Loader)
-    config.update(vars(args))
-
-    # check arguments
-    if (args.wav_scp is not None and args.rootdir is not None) or (
-        args.wav_scp is None and args.rootdir is None
-    ):
-        raise ValueError("Please specify either --rootdir or --wav-scp.")
-
-    # get dataset
-    if args.rootdir is not None:
-        dataset = AudioDataset(
-            args.rootdir,
-            "*.wav",
-            audio_load_fn=sf.read,
-            return_utt_id=True,
-        )
-    else:
-        dataset = AudioSCPDataset(
-            args.wav_scp,
-            segments=args.segments,
-            return_utt_id=True,
-            return_sampling_rate=True,
-        )
-
-    # get text
-    with open(args.text) as f:
-        lines = [line.strip() for line in f.readlines()]
-    text = {
-        line.split(maxsplit=1)[0]: line.split(maxsplit=1)[1].split() for line in lines
-    }
-
-    # load spk2utt file
-    if args.utt2spk is not None:
-        with open(args.utt2spk) as f:
-            lines = [l.replace("\n", "") for l in f.readlines()]
-        utt2spk = {l.split()[0]: l.split()[1] for l in lines}
-        with open(args.spk2idx) as f:
-            lines = [l.replace("\n", "") for l in f.readlines()]
-        spk2idx = {l.split()[0]: int(l.split()[1]) for l in lines}
-
-    # check directly existence
-    if not os.path.exists(args.dumpdir):
-        os.makedirs(args.dumpdir, exist_ok=True)
-
-    # process each data
-    for utt_id, (audio, fs) in tqdm(dataset):
-        # check
-        assert len(audio.shape) == 1, f"{utt_id} seems to be multi-channel signal."
-        assert (
-            np.abs(audio).max() <= 1.0
-        ), f"{utt_id} seems to be different from 16 bit PCM."
-
-        # downsample
-        if fs != config["sampling_rate"]:
-            audio = resampy.resample(audio, fs, config["sampling_rate"], axis=0)
-
-        # trim silence
-        if config["trim_silence"]:
-            audio, _ = librosa.effects.trim(
-                audio,
-                top_db=config["trim_threshold_in_db"],
-                frame_length=config["trim_frame_size"],
-                hop_length=config["trim_hop_size"],
-            )
-
-        # use hubert index instead of mel
-        mel = np.array(text[utt_id]).astype(np.int64).reshape(-1, 1)
-
-        if args.spk2idx is not None:
-            spk = utt2spk[utt_id]
-            if spk in spk2idx:
-                idx = spk2idx[spk]
-            else:
-                logging.warn(f"{spk} is unknown speaker.")
-                max_idx = max(spk2idx.values()) + 1
-                idx = max_idx
-
-            # concatenate with mel
-            idx = np.repeat(np.array(idx).reshape(1, 1), len(mel), axis=0)
-            mel = np.concatenate([mel, idx], axis=1)
-
-        # make sure the audio length and feature length are matched
-        logging.info(f"Mod: {len(audio) - len(mel) * config['hop_size']}")
-        if len(mel) * config["hop_size"] <= len(audio):
-            logging.warning(
-                "len(mel) * config['hop_size'] <= len(audio), may be errors"
-            )
-        mel = mel[: len(audio) // config["hop_size"]]
-        audio = audio[: len(mel) * config["hop_size"]]
-        assert len(mel) * config["hop_size"] == len(audio)
-
-        # apply global gain
-        if config["global_gain_scale"] > 0.0:
-            audio *= config["global_gain_scale"]
-        if np.abs(audio).max() >= 1.0:
-            logging.warn(
-                f"{utt_id} causes clipping. "
-                "it is better to re-consider global gain scale."
-            )
-            continue
-
-        # save
-        if config["format"] == "hdf5":
-            write_hdf5(
-                os.path.join(args.dumpdir, f"{utt_id}.h5"),
-                "wave",
-                audio.astype(np.float32),
-            )
-            write_hdf5(
-                os.path.join(args.dumpdir, f"{utt_id}.h5"),
-                "feats",
-                mel.astype(np.float32),
-            )
-        elif config["format"] == "npy":
-            np.save(
-                os.path.join(args.dumpdir, f"{utt_id}-wave.npy"),
-                audio.astype(np.float32),
-                allow_pickle=False,
-            )
-            np.save(
-                os.path.join(args.dumpdir, f"{utt_id}-feats.npy"),
-                mel.astype(np.float32),
-                allow_pickle=False,
-            )
-        else:
-            raise ValueError("support only hdf5 or npy format.")
-
-
-if __name__ == "__main__":
-    main()

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/path.sh

@@ -1,33 +0,0 @@
-# cuda related
-export CUDA_HOME=/usr/local/cuda-11.3
-export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
-
-# path related
-export PRJ_ROOT="${PWD}/../../.."
-if [ -e "${PRJ_ROOT}/tools/venv/bin/activate" ]; then
-    # shellcheck disable=SC1090
-    . "${PRJ_ROOT}/tools/venv/bin/activate"
-fi
-
-# python related
-export OMP_NUM_THREADS=1
-export PYTHONIOENCODING=UTF-8
-export MPL_BACKEND=Agg
-
-# check installation
-if ! command -v parallel-wavegan-train > /dev/null; then
-    echo "Error: It seems setup is not finished." >&2
-    echo "Error: Please setup your environment by following README.md" >&2
-    return 1
-fi
-if ! command -v jq > /dev/null; then
-    echo "Error: It seems jq is not installed." >&2
-    echo "Error: Please install via \`sudo apt-get install jq\`." >&2
-    echo "Error: If you do not have sudo, please download from https://stedolan.github.io/jq/download/." >&2
-    return 1
-fi
-if ! command -v yq > /dev/null; then
-    echo "Error: It seems yq is not installed." >&2
-    echo "Error: Please install via \`pip install yq\`." >&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/voc1/run.sh

@@ -1,188 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-. ./cmd.sh || exit 1;
-. ./path.sh || exit 1;
-
-# basic settings
-stage=-1       # stage to start
-stop_stage=100 # stage to stop
-verbose=1      # verbosity level (lower is less info)
-n_gpus=1       # number of gpus in training
-n_jobs=16      # number of parallel jobs in feature extraction
-
-# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
-conf=conf/parallel_wavegan.v1.yaml
-
-# speaker setting
-spks="all" # all or you can choose speakers e.g., "p225 p226 p227 ..."
-
-# directory path setting
-download_dir=downloads # directory to save database
-dumpdir=dump           # directory to dump features
-
-# training related setting
-tag=""     # tag for directory to save model
-resume=""  # checkpoint path to resume training
-           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
-
-# decoding related setting
-checkpoint="" # checkpoint path to be used for decoding
-              # if not provided, the latest one will be used
-              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-train_set="train_nodev_$(echo "${spks}" | tr " " "_")" # name of training data directory
-dev_set="dev_$(echo "${spks}" | tr " " "_")"           # name of development data directory
-eval_set="eval_$(echo "${spks}" | tr " " "_")"         # name of evaluation data directory
-
-set -euo pipefail
-
-if [ "${stage}" -le -1 ] && [ "${stop_stage}" -ge -1 ]; then
-    echo "Stage -1: Data download"
-    local/data_download.sh "${download_dir}"
-fi
-
-if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
-    echo "Stage 0: Data preparation"
-    train_data_dirs=""
-    dev_data_dirs=""
-    eval_data_dirs=""
-    # if set to "all", use all of the speakers in the corpus
-    if [ "${spks}" = "all" ]; then
-        # NOTE(kan-bayashi): p315 will not be used since it lacks txt data
-        spks=$(find "${download_dir}/VCTK-Corpus/wav48" \
-            -maxdepth 1 -name "p*" -exec basename {} \; | sort | grep -v p315)
-    fi
-    for spk in ${spks}; do
-        local/data_prep.sh \
-            --fs "$(yq ".sampling_rate" "${conf}")" \
-            --train_set "train_nodev_${spk}" \
-            --dev_set "dev_${spk}" \
-            --eval_set "eval_${spk}" \
-            "${download_dir}/VCTK-Corpus" "${spk}" data
-        train_data_dirs+=" data/train_nodev_${spk}"
-        dev_data_dirs+=" data/dev_${spk}"
-        eval_data_dirs+=" data/eval_${spk}"
-    done
-    # shellcheck disable=SC2086
-    utils/combine_data.sh "data/${train_set}" ${train_data_dirs}
-    # shellcheck disable=SC2086
-    utils/combine_data.sh "data/${dev_set}" ${dev_data_dirs}
-    # shellcheck disable=SC2086
-    utils/combine_data.sh "data/${eval_set}" ${eval_data_dirs}
-fi
-
-stats_ext=$(grep -q "hdf5" <(yq ".format" "${conf}") && echo "h5" || echo "npy")
-if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
-    echo "Stage 1: Feature extraction"
-    # extract raw features
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
-        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
-        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
-            parallel-wavegan-preprocess \
-                --config "${conf}" \
-                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
-                --segments "${dumpdir}/${name}/raw/segments.JOB" \
-                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished feature extraction of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished feature extraction."
-
-    # calculate statistics for normalization
-    echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics.log."
-    ${train_cmd} "${dumpdir}/${train_set}/compute_statistics.log" \
-        parallel-wavegan-compute-statistics \
-            --config "${conf}" \
-            --rootdir "${dumpdir}/${train_set}/raw" \
-            --dumpdir "${dumpdir}/${train_set}" \
-            --verbose "${verbose}"
-    echo "Successfully finished calculation of statistics."
-
-    # normalize and dump them
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/norm" ] && mkdir -p "${dumpdir}/${name}/norm"
-        echo "Nomalization start. See the progress via ${dumpdir}/${name}/norm/normalize.*.log."
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/norm/normalize.JOB.log" \
-            parallel-wavegan-normalize \
-                --config "${conf}" \
-                --stats "${dumpdir}/${train_set}/stats.${stats_ext}" \
-                --rootdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --dumpdir "${dumpdir}/${name}/norm/dump.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished normalization of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished normalization."
-fi
-
-if [ -z "${tag}" ]; then
-    expdir="exp/${train_set}_vctk_$(basename "${conf}" .yaml)"
-else
-    expdir="exp/${train_set}_vctk_${tag}"
-fi
-if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
-    echo "Stage 2: Network training"
-    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
-    cp "${dumpdir}/${train_set}/stats.${stats_ext}" "${expdir}"
-    if [ "${n_gpus}" -gt 1 ]; then
-        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
-    else
-        train="parallel-wavegan-train"
-    fi
-    echo "Training start. See the progress via ${expdir}/train.log."
-    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
-        ${train} \
-            --config "${conf}" \
-            --train-dumpdir "${dumpdir}/${train_set}/norm" \
-            --dev-dumpdir "${dumpdir}/${dev_set}/norm" \
-            --outdir "${expdir}" \
-            --resume "${resume}" \
-            --verbose "${verbose}"
-    echo "Successfully finished training."
-fi
-
-if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
-    echo "Stage 3: Network decoding"
-    # shellcheck disable=SC2012
-    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
-        [ "${n_gpus}" -gt 1 ] && n_gpus=1
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
-        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
-            parallel-wavegan-decode \
-                --dumpdir "${dumpdir}/${name}/norm" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished decoding."
-fi
-echo "Finished."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/cmd.sh

@@ -1,91 +0,0 @@
-# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
-# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
-# e.g.
-#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
-#
-# Options:
-#   --time <time>: Limit the maximum time to execute.
-#   --mem <mem>: Limit the maximum memory usage.
-#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
-#   --num-threads <ngpu>: Specify the number of CPU core.
-#   --gpu <ngpu>: Specify the number of GPU devices.
-#   --config: Change the configuration file from default.
-#
-# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
-# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
-# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
-# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
-#
-# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
-# These options are mapping to specific options for each backend and
-# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
-# If jobs failed, your configuration might be wrong for your environment.
-#
-#
-# The official documentaion for run.pl, queue.pl, slurm.pl, and ssh.pl:
-#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
-# =========================================================~
-
-
-# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
-cmd_backend="local"
-
-# Local machine, without any Job scheduling system
-if [ "${cmd_backend}" = local ]; then
-
-    # The other usage
-    export train_cmd="utils/run.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/run.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/run.pl"
-
-# Local machine, without any Job scheduling system
-elif [ "${cmd_backend}" = stdout ]; then
-
-    # The other usage
-    export train_cmd="utils/stdout.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/stdout.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/stdout.pl"
-
-# "qsub" (SGE, Torque, PBS, etc.)
-elif [ "${cmd_backend}" = sge ]; then
-    # The default setting is written in conf/queue.conf.
-    # You must change "-q g.q" for the "queue" for your environment.
-    # To know the "queue" names, type "qhost -q"
-    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
-
-    export train_cmd="utils/queue.pl"
-    export cuda_cmd="utils/queue.pl"
-    export decode_cmd="utils/queue.pl"
-
-# "sbatch" (Slurm)
-elif [ "${cmd_backend}" = slurm ]; then
-    # The default setting is written in conf/slurm.conf.
-    # You must change "-p cpu" and "-p gpu" for the "partion" for your environment.
-    # To know the "partion" names, type "sinfo".
-    # You can use "--gpu * " by defualt for slurm and it is interpreted as "--gres gpu:*"
-    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
-
-    export train_cmd="utils/slurm.pl"
-    export cuda_cmd="utils/slurm.pl"
-    export decode_cmd="utils/slurm.pl"
-
-elif [ "${cmd_backend}" = ssh ]; then
-    # You have to create ".queue/machines" to specify the host to execute jobs.
-    # e.g. .queue/machines
-    #   host1
-    #   host2
-    #   host3
-    # Assuming you can login them without any password, i.e. You have to set ssh keys.
-
-    export train_cmd="utils/ssh.pl"
-    export cuda_cmd="utils/ssh.pl"
-    export decode_cmd="utils/ssh.pl"
-
-else
-    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/conf/conditioned_melgan_vae.v3.yaml

@@ -1,130 +0,0 @@
-###########################################################
-#    THIS IS EXPERIMENTAL CONFIGURATION! BE CAREFUL!      #
-###########################################################
-# This is the hyperparameter configuration file for VQ-VAE MelGAN.
-# Please make sure this is adjusted for the LJSpeech dataset. If you want to
-# apply to the other dataset, you might need to carefully change some parameters.
-# This configuration performs 5M iters and it requires ~4 GB GPU memory and
-# takes ~8 days on TITAN V.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000       # Sampling rate.
-global_gain_scale: 1.0     # Will be multiplied to all of waveform.
-trim_silence: true         # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20   # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024      # Frame size in trimming.
-trim_hop_size: 256         # Hop size in trimming.
-use_global_condition: true # Whether to use global conditioning.
-format: "hdf5"             # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "VQVAE" # Generator type.
-generator_params:
-    in_channels: 1                      # Number of input channels.
-    out_channels: 1                     # Number of output channels.
-    num_embeds: 512                     # Number of embeddings.
-    embed_dim: 256                      # Dimension of each emebedding.
-    num_global_embeds: 128              # Number of global embeddings.
-    global_embed_dim: 128               # Dimension of each global embedding.
-    encoder_type: "MelGANDiscriminator" # Encoder type.
-    decoder_type: "MelGANGenerator"     # Decoder type.
-    encoder_conf:
-        "out_channels": 256 # Must be the same as embed dim
-        "downsample_scales": [4, 4, 2, 2]
-        "max_downsample_channels": 1024
-    decoder_conf:
-        "in_channels": 384 # Must be the same as embed dim
-        "upsample_scales": [4, 4, 2, 2]
-        "channels": 512
-        "stacks": 3
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
-discriminator_params:
-    in_channels: 1                    # Number of input channels.
-    out_channels: 1                   # Number of output channels.
-    scales: 3                         # Number of multi-scales.
-    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
-    downsample_pooling_params:        # Parameters of the above pooling function.
-        kernel_size: 4
-        stride: 2
-        padding: 1
-        count_include_pad: False
-    kernel_sizes: [5, 3]              # List of kernel size.
-    channels: 16                      # Number of channels of the initial conv layer.
-    max_downsample_channels: 1024     # Maximum number of channels of downsampling layers.
-    downsample_scales: [4, 4, 4, 4]   # List of downsampling scales.
-    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
-    nonlinear_activation_params:      # Parameters of nonlinear activation function.
-        negative_slope: 0.2
-    use_weight_norm: True             # Whether to use weight norm.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-use_feat_match_loss: true
-lambda_commit: 0.25
-lambda_feat_match: 25.0
-lambda_adv: 4.0
-lambda_aux_after_introduce_adv_loss: 1.0
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 16             # Batch size.
-batch_max_steps: 8192      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2             # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 5000000     # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 5000000     # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
-train_max_steps: 5000000                # Number of training steps.
-save_interval_steps: 5000               # Interval steps to save checkpoint.
-eval_interval_steps: 1000               # Interval steps to evaluate the network.
-log_interval_steps: 100                 # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/conf/local_conditioned_melgan_vae.v3.yaml

@@ -1,134 +0,0 @@
-###########################################################
-#    THIS IS EXPERIMENTAL CONFIGURATION! BE CAREFUL!      #
-###########################################################
-# This is the hyperparameter configuration file for VQ-VAE MelGAN.
-# Please make sure this is adjusted for the LJSpeech dataset. If you want to
-# apply to the other dataset, you might need to carefully change some parameters.
-# This configuration performs 5M iters and it requires ~4 GB GPU memory and
-# takes ~8 days on TITAN V.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 24000       # Sampling rate.
-hop_size: 64               # Hop size.
-global_gain_scale: 1.0     # Will be multiplied to all of waveform.
-trim_silence: true         # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20   # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024      # Frame size in trimming.
-trim_hop_size: 256         # Hop size in trimming.
-use_global_condition: true # Whether to use global conditioning.
-use_local_condition: true  # Whether to use local conditioning.
-format: "hdf5"             # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "VQVAE" # Generator type.
-generator_params:
-    in_channels: 1                      # Number of input channels.
-    out_channels: 1                     # Number of output channels.
-    num_embeds: 512                     # Number of embeddings.
-    embed_dim: 256                      # Dimension of each emebedding.
-    num_local_embeds: 2                 # Number of local embeddings.
-    local_embed_dim: 32                 # Dimension of each local embedding.
-    num_global_embeds: 128              # Number of global embeddings.
-    global_embed_dim: 128               # Dimension of each global embedding.
-    encoder_type: "MelGANDiscriminator" # Encoder type.
-    decoder_type: "MelGANGenerator"     # Decoder type.
-    encoder_conf:
-        "out_channels": 256 # Must be the same as embed dim
-        "downsample_scales": [4, 4, 2, 2]
-        "max_downsample_channels": 1024
-    decoder_conf:
-        "in_channels": 416 # Must be the same as embed + local embed + global embed dim
-        "upsample_scales": [4, 4, 2, 2]
-        "channels": 512
-        "stacks": 3
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
-discriminator_params:
-    in_channels: 1                    # Number of input channels.
-    out_channels: 1                   # Number of output channels.
-    scales: 3                         # Number of multi-scales.
-    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
-    downsample_pooling_params:        # Parameters of the above pooling function.
-        kernel_size: 4
-        stride: 2
-        padding: 1
-        count_include_pad: False
-    kernel_sizes: [5, 3]              # List of kernel size.
-    channels: 16                      # Number of channels of the initial conv layer.
-    max_downsample_channels: 1024     # Maximum number of channels of downsampling layers.
-    downsample_scales: [4, 4, 4, 4]   # List of downsampling scales.
-    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
-    nonlinear_activation_params:      # Parameters of nonlinear activation function.
-        negative_slope: 0.2
-    use_weight_norm: True             # Whether to use weight norm.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-use_feat_match_loss: true
-lambda_commit: 0.25
-lambda_feat_match: 25.0
-lambda_adv: 4.0
-lambda_aux_after_introduce_adv_loss: 1.0
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 16             # Batch size.
-batch_max_steps: 8192      # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true           # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2             # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true          # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 5000000     # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 5000000     # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 100000 # Number of steps to start to train discriminator.
-train_max_steps: 5000000                # Number of training steps.
-save_interval_steps: 5000               # Interval steps to save checkpoint.
-eval_interval_steps: 1000               # Interval steps to evaluate the network.
-log_interval_steps: 100                 # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/conf/slurm.conf

@@ -1,12 +0,0 @@
-# Default configuration
-command sbatch --export=PATH  --ntasks-per-node=1
-option time=* --time $0
-option mem=* --mem-per-cpu $0
-option mem=0          # Do not add anything to qsub_opts
-option num_threads=* --cpus-per-task $0 --ntasks-per-node=1
-option num_threads=1 --cpus-per-task 1  --ntasks-per-node=1 # Do not add anything to qsub_opts
-default gpu=0
-option gpu=0 -p cpu
-option gpu=* -p gpu --gres=gpu:$0
-# note: the --max-jobs-run option is supported as a special case
-# by slurm.pl and you don't have to handle it in the config file.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/local/data_download.sh

@@ -1,37 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-download_dir=$1
-
-# check arguments
-if [ $# != 1 ]; then
-    echo "Usage: $0 <download_dir>"
-    exit 1
-fi
-
-set -euo pipefail
-
-cwd=$(pwd)
-if [ ! -e "${download_dir}/VCTK-Corpus" ]; then
-    mkdir -p "${download_dir}"
-    cd "${download_dir}" || exit 1;
-    wget http://www.udialogue.org/download/VCTK-Corpus.tar.gz
-    tar xvzf ./VCTK-Corpus.tar.gz
-    rm ./VCTK-Corpus.tar.gz
-    cd "${cwd}" || exit 1;
-    echo "Successfully downloaded data."
-else
-    echo "Already exists. Skipped."
-fi
-
-if [ ! -e "${download_dir}/VCTK-Corpus/lab" ]; then
-    cd "${download_dir}" || exit 1;
-    git clone https://github.com/kan-bayashi/VCTKCorpusFullContextLabel.git
-    cp -r VCTKCorpusFullContextLabel/lab ./VCTK-Corpus
-    cd "${cwd}" || exit 1;
-    echo "Successfully downloaded label data."
-else
-    echo "Already exists. Skipped."
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/local/data_prep.sh

@@ -1,121 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-# shellcheck disable=SC1091
-. ./path.sh || exit 1;
-
-fs=24000
-num_dev=10
-num_eval=10
-train_set="train_nodev"
-dev_set="dev"
-eval_set="eval"
-shuffle=false
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-db_root=$1
-spk=$2
-data_dir=$3
-
-# check arguments
-if [ $# != 3 ]; then
-    echo "Usage: $0 [Options] <db_root> <spk> <data_dir>"
-    echo "e.g.: $0 downloads/VCTK-Corpus p225 data"
-    echo ""
-    echo "Options:"
-    echo "    --fs: sampling frequency (default=24000)."
-    echo "    --num_dev: number of development uttreances (default=10)."
-    echo "    --num_eval: number of evaluation uttreances (default=10)."
-    echo "    --train_set: name of train set (default=train_nodev)."
-    echo "    --dev_set: name of dev set (default=dev)."
-    echo "    --eval_set: name of eval set (default=eval)."
-    echo "    --shuffle: whether to perform shuffle in making dev / eval set (default=false)."
-    exit 1
-fi
-
-set -euo pipefail
-
-# check spk existence
-[ ! -e "${db_root}/lab/mono/${spk}" ] && \
-    echo "${spk} does not exist." >&2 && exit 1;
-
-[ ! -e "${data_dir}/all" ] && mkdir -p "${data_dir}/all"
-
-# set filenames
-scp="${data_dir}/all/wav.scp"
-segments="${data_dir}/all/segments"
-utt2spk="${data_dir}/all/utt2spk"
-
-# check file existence
-[ -e "${scp}" ] && rm "${scp}"
-[ -e "${segments}" ] && rm "${segments}"
-[ -e "${utt2spk}" ] && rm "${utt2spk}"
-
-# make scp, utt2spk, and segments
-find "${db_root}/wav48/${spk}" -follow -name "*.wav" | sort | while read -r wav; do
-    id=$(basename "${wav}" | sed -e "s/\.[^\.]*$//g")
-    lab=${db_root}/lab/mono/${spk}/${id}.lab
-
-    # check lab existence
-    if [ ! -e "${lab}" ]; then
-        echo "${id} has not label file. skipped."
-        continue
-    fi
-
-    echo "${id} cat ${wav} | sox -t wav - -c 1 -b 16 -t wav - rate ${fs} |" >> "${scp}"
-    echo "${id} ${spk}" >> "${utt2spk}"
-
-    # parse start and end time from HTS-style mono label
-    idx=1
-    while true; do
-        next_idx=$((idx+1))
-        next_symbol=$(sed -n "${next_idx}p" "${lab}" | awk '{print $3}')
-        if [ "${next_symbol}" != "pau" ]; then
-            start_nsec=$(sed -n "${idx}p" "${lab}" | awk '{print $2}')
-            break
-        fi
-        idx=${next_idx}
-    done
-    idx=$(wc -l < "${lab}")
-    while true; do
-        prev_idx=$((idx-1))
-        prev_symbol=$(sed -n "${prev_idx}p" "${lab}" | awk '{print $3}')
-        if [ "${prev_symbol}" != "pau" ]; then
-            end_nsec=$(sed -n "${idx}p" "${lab}" | awk '{print $1}')
-            break
-        fi
-        idx=${prev_idx}
-    done
-    start_sec=$(echo "${start_nsec}*0.0000001" | bc | sed "s/^\./0./")
-    end_sec=$(echo "${end_nsec}*0.0000001" | bc | sed "s/^\./0./")
-    echo "${id} ${id} ${start_sec} ${end_sec}" >> "${segments}"
-done
-
-# split
-num_all=$(wc -l < "${scp}")
-num_deveval=$((num_dev + num_eval))
-num_train=$((num_all - num_deveval))
-utils/split_data.sh \
-    --num_first "${num_train}" \
-    --num_second "${num_deveval}" \
-    --shuffle "${shuffle}" \
-    "${data_dir}/all" \
-    "${data_dir}/${train_set}" \
-    "${data_dir}/deveval"
-utils/split_data.sh \
-    --num_first "${num_dev}" \
-    --num_second "${num_eval}" \
-    --shuffle "${shuffle}" \
-    "${data_dir}/deveval" \
-    "${data_dir}/${dev_set}" \
-    "${data_dir}/${eval_set}"
-
-# remove tmp directories
-rm -rf "${data_dir}/all"
-rm -rf "${data_dir}/deveval"
-
-echo "Successfully prepared data."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/local/data_prep_2mix.sh

@@ -1,54 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-# shellcheck disable=SC1091
-. ./path.sh || exit 1;
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-db_root=$1
-data_dir_root=$2
-
-# check arguments
-if [ $# != 2 ]; then
-    echo "Usage: $0 <db_root> <data_dir>"
-    echo "e.g.: $0 downloads/VCTK-Corpus data"
-    exit 1
-fi
-
-set -euo pipefail
-
-for mix_type in min max; do
-    for dset in tr tt cv; do
-        for src_spk in s1 s2; do
-            echo "${mix_type}_${dset}_${src_spk}"
-            # set filenames
-            data_dir=${data_dir_root}/${mix_type}_${dset}_${src_spk}
-            scp="${data_dir}/wav.scp"
-            utt2spk="${data_dir}/utt2spk"
-
-            # check file existence
-            [ ! -e "${data_dir}" ] && mkdir -p "${data_dir}"
-            [ -e "${scp}" ] && rm "${scp}"
-            [ -e "${utt2spk}" ] && rm "${utt2spk}"
-
-            # make scp, utt2spk, and segments
-            find "${db_root}/${mix_type}/${dset}/${src_spk}" -follow -name "*.wav" | sort | while read -r wav; do
-                id=$(basename "${wav}" | sed -e "s/\.[^\.]*$//g")
-                if [ ${src_spk} = "s1" ]; then
-                    spk=$(echo "${id}" | cut -f 1 -d "_")
-                else
-                    spk=$(echo "${id}" | cut -f 4 -d "_")
-                fi
-                echo "${id} ${wav}" >> "${scp}"
-                echo "${id} ${spk}" >> "${utt2spk}"
-            done
-        done
-    done
-done
-
-
-echo "Successfully prepared data."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/local/decode_from_text.py

@@ -1,174 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-"""Decode text with trained VQ-VAE Generator or discrete symbol vocoder."""
-
-import argparse
-import logging
-import os
-import time
-
-import parallel_wavegan.models
-import soundfile as sf
-import torch
-import yaml
-from parallel_wavegan.utils import load_model
-from tqdm import tqdm
-
-
-def main():
-    """Run decoding process."""
-    parser = argparse.ArgumentParser(
-        description=(
-            "Decode text with trained VQ-VAE decoder "
-            "(See detail in parallel_wavegan/bin/decode.py)."
-        )
-    )
-    parser.add_argument(
-        "--text",
-        required=True,
-        type=str,
-        help="kaldi-style text file.",
-    )
-    parser.add_argument(
-        "--utt2spk",
-        default=None,
-        type=str,
-        help="kaldi-style utt2spk file.",
-    )
-    parser.add_argument(
-        "--spk2idx",
-        default=None,
-        type=str,
-        help="kaldi-style spk2idx file.",
-    )
-    parser.add_argument(
-        "--outdir",
-        type=str,
-        required=True,
-        help="directory to save generated speech.",
-    )
-    parser.add_argument(
-        "--checkpoint",
-        type=str,
-        required=True,
-        help="checkpoint file to be loaded.",
-    )
-    parser.add_argument(
-        "--config",
-        default=None,
-        type=str,
-        help=(
-            "yaml format configuration file. if not explicitly provided, "
-            "it will be searched in the checkpoint directory. (default=None)"
-        ),
-    )
-    parser.add_argument(
-        "--verbose",
-        type=int,
-        default=1,
-        help="logging level. higher is more logging. (default=1)",
-    )
-    args = parser.parse_args()
-
-    # set logger
-    if args.verbose > 1:
-        logging.basicConfig(
-            level=logging.DEBUG,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    elif args.verbose > 0:
-        logging.basicConfig(
-            level=logging.INFO,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    else:
-        logging.basicConfig(
-            level=logging.WARN,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-        logging.warning("Skip DEBUG/INFO messages")
-
-    # check directory existence
-    if not os.path.exists(args.outdir):
-        os.makedirs(args.outdir)
-
-    # load config
-    if args.config is None:
-        dirname = os.path.dirname(args.checkpoint)
-        args.config = os.path.join(dirname, "config.yml")
-    with open(args.config) as f:
-        config = yaml.load(f, Loader=yaml.Loader)
-    config.update(vars(args))
-
-    # setup model
-    if torch.cuda.is_available():
-        device = torch.device("cuda")
-    else:
-        device = torch.device("cpu")
-    model = load_model(args.checkpoint, config)
-    logging.info(f"Loaded model parameters from {args.checkpoint}.")
-    model.remove_weight_norm()
-    model = model.eval().to(device)
-    is_vqvae = isinstance(model, parallel_wavegan.models.VQVAE)
-
-    # setup dataset
-    with open(args.text) as f:
-        lines = [l_.replace("\n", "") for l_ in f.readlines()]
-    text = {l_.split()[0]: list(map(int, l_.split()[1:])) for l_ in lines}
-
-    utt2spk = None
-    if args.utt2spk is not None:
-        assert args.spk2idx is not None
-        with open(args.utt2spk) as f:
-            lines = [l_.replace("\n", "") for l_ in f.readlines()]
-        utt2spk = {l_.split()[0]: str(l_.split()[1]) for l_ in lines}
-        with open(args.spk2idx) as f:
-            lines = [l_.replace("\n", "") for l_ in f.readlines()]
-        spk2idx = {l_.split()[0]: int(l_.split()[1]) for l_ in lines}
-
-    # start generation
-    total_rtf = 0.0
-    with torch.no_grad(), tqdm(text.items(), desc="[decode]") as pbar:
-        for idx, items in enumerate(pbar, 1):
-            utt_id, indices = items
-            z = torch.LongTensor(indices).view(1, -1).to(device)
-            g = None
-            if utt2spk is not None:
-                spk_idx = spk2idx[utt2spk[utt_id]]
-                g = torch.tensor(spk_idx).long().view(1).to(device)
-            if is_vqvae:
-                # VQVAE case
-                start = time.time()
-                y = model.decode(z, None, g).view(-1).cpu().numpy()
-                rtf = (time.time() - start) / (len(y) / config["sampling_rate"])
-                pbar.set_postfix({"RTF": rtf})
-                total_rtf += rtf
-            else:
-                # Discrete symbol vocoder case
-                start = time.time()
-                g = int(g.item()) if g is not None else None
-                y = model.inference(z.view(-1, 1), g=g).view(-1).cpu().numpy()
-                rtf = (time.time() - start) / (len(y) / config["sampling_rate"])
-                pbar.set_postfix({"RTF": rtf})
-                total_rtf += rtf
-
-            # save as PCM 16 bit wav file
-            sf.write(
-                os.path.join(config["outdir"], f"{utt_id}_gen.wav"),
-                y,
-                config["sampling_rate"],
-                "PCM_16",
-            )
-
-    # report average RTF
-    logging.info(
-        f"Finished generation of {idx} utterances (RTF = {total_rtf / idx:.03f})."
-    )
-
-
-if __name__ == "__main__":
-    main()

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/local/setup_for_vc.py

@@ -1,222 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-"""Setup features for voice conversion."""
-
-import argparse
-import logging
-import os
-
-import numpy as np
-import yaml
-from parallel_wavegan.datasets import AudioDataset
-from parallel_wavegan.utils import read_hdf5, write_hdf5
-from tqdm import tqdm
-
-
-def main():
-    """Run preprocessing process."""
-    parser = argparse.ArgumentParser(
-        description=(
-            "Normalize dumped raw features (See detail in"
-            " parallel_wavegan/bin/normalize.py)."
-        )
-    )
-    parser.add_argument(
-        "--rootdir",
-        default=None,
-        type=str,
-        help=(
-            "directory including feature files to be normalized. "
-            "you need to specify either feats-scp or rootdir."
-        ),
-    )
-    parser.add_argument(
-        "--feats-scp",
-        default=None,
-        type=str,
-        help=(
-            "kaldi-style feats.scp file. "
-            "you need to specify either feats-scp or rootdir."
-        ),
-    )
-    parser.add_argument(
-        "--dumpdir", type=str, required=True, help="directory to dump feature files."
-    )
-    parser.add_argument(
-        "--statdir",
-        type=str,
-        default=None,
-        help="direcotry including statistics files.",
-    )
-    parser.add_argument(
-        "--utt2spk", default=None, type=str, help="kaldi-style utt2spk file."
-    )
-    parser.add_argument(
-        "--spk2idx", default=None, type=str, help="kaldi-style spk2idx file."
-    )
-    parser.add_argument(
-        "--config", type=str, required=True, help="yaml format configuration file."
-    )
-    parser.add_argument(
-        "--verbose",
-        type=int,
-        default=1,
-        help="logging level. higher is more logging. (default=1)",
-    )
-    args = parser.parse_args()
-
-    # set logger
-    if args.verbose > 1:
-        logging.basicConfig(
-            level=logging.DEBUG,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    elif args.verbose > 0:
-        logging.basicConfig(
-            level=logging.INFO,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-    else:
-        logging.basicConfig(
-            level=logging.WARN,
-            format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
-        )
-        logging.warning("Skip DEBUG/INFO messages")
-
-    # load config
-    with open(args.config) as f:
-        config = yaml.load(f, Loader=yaml.Loader)
-    config.update(vars(args))
-
-    # check arguments
-    if (args.feats_scp is not None and args.rootdir is not None) or (
-        args.feats_scp is None and args.rootdir is None
-    ):
-        raise ValueError("Please specify either --rootdir or --feats-scp.")
-
-    # check directory existence
-    if not os.path.exists(args.dumpdir):
-        os.makedirs(args.dumpdir)
-
-    # load utt2spk and spk2idx
-    with open(args.utt2spk) as f:
-        lines = [l_.replace("\n", "") for l_ in f.readlines()]
-    utt2spk = {l_.split()[0]: l_.split()[1] for l_ in lines}
-    with open(args.spk2idx) as f:
-        lines = [l_.replace("\n", "") for l_ in f.readlines()]
-    spk2idx = {l_.split()[0]: int(l_.split()[1]) for l_ in lines}
-    spks = list(spk2idx.keys())
-
-    # get dataset
-    if args.rootdir is not None:
-        local_query = None
-        local_load_fn = None
-        if config["format"] == "hdf5":
-            audio_query = "*.h5"
-            audio_load_fn = lambda x: read_hdf5(x, "wave")  # NOQA
-            if config.get("use_local_condition", False):
-                local_query = "*.h5"
-                local_load_fn = lambda x: read_hdf5(x, "local")  # NOQA
-        elif config["format"] == "npy":
-            audio_query = "*-wave.npy"
-            audio_load_fn = np.load
-            if config.get("use_local_condition", False):
-                local_query = "*-local.npy"
-                local_load_fn = np.load
-        else:
-            raise ValueError("support only hdf5 or npy format.")
-        dataset = AudioDataset(
-            root_dir=args.rootdir,
-            audio_query=audio_query,
-            audio_load_fn=audio_load_fn,
-            local_query=local_query,
-            local_load_fn=local_load_fn,
-            return_utt_id=True,
-        )
-    else:
-        raise NotImplementedError("Not supported.")
-    logging.info(f"The number of files = {len(dataset)}.")
-
-    # load statistics
-    if config.get("use_local_condition", False):
-        spk2lf0avg = {}
-        for spk in spks:
-            # FIXME: it is better to remove this hard coding
-            if config["format"] == "hdf5":
-                lf0avg = read_hdf5(f"{args.statdir}/stats.h5", f"{spk}/mean")[0]
-            elif config["format"] == "npy":
-                lf0avg = np.load(f"{args.statdir}/stats-${spk}.h5")[0][0]
-            else:
-                raise ValueError("support only hdf5 or npy format.")
-            spk2lf0avg[spk] = lf0avg
-
-    # process each file
-    for items in tqdm(dataset):
-        if config.get("use_local_condition", False):
-            utt_id, audio, l_ = items
-        else:
-            utt_id, audio = items
-        source_spk = utt2spk[utt_id]
-
-        # replace global and adjust f0 range
-        for target_spk in spks:
-            if target_spk == source_spk:
-                continue
-
-            # adjust f0 range
-            if config.get("use_local_condition", False):
-                l_ = l_.copy()
-                target_lf0avg = spk2lf0avg[target_spk]
-                source_lf0avg = spk2lf0avg[source_spk]
-                l_[:, 0] = l_[:, 0] + target_lf0avg - source_lf0avg
-
-            # get global
-            g = np.array(spk2idx[target_spk])
-
-            # save
-            if config["format"] == "hdf5":
-                write_hdf5(
-                    os.path.join(args.dumpdir, f"{utt_id}_to_{target_spk}.h5"),
-                    "wave",
-                    audio.astype(np.float32),
-                )
-                write_hdf5(
-                    os.path.join(args.dumpdir, f"{utt_id}_to_{target_spk}.h5"),
-                    "global",
-                    g.reshape(-1),
-                )
-                if config.get("use_local_condition", False):
-                    write_hdf5(
-                        os.path.join(args.dumpdir, f"{utt_id}_to_{target_spk}.h5"),
-                        "local",
-                        l_.astype(np.float32),
-                    )
-            elif config["format"] == "npy":
-                np.save(
-                    os.path.join(args.dumpdir, f"{utt_id}_to_{target_spk}-wave.npy"),
-                    audio.astype(np.float32),
-                    allow_pickle=False,
-                )
-                np.save(
-                    os.path.join(args.dumpdir, f"{utt_id}_to_{target_spk}-global.npy"),
-                    g.reshape(-1),
-                    allow_pickle=False,
-                )
-                if config.get("use_local_condition", False):
-                    np.save(
-                        os.path.join(
-                            args.dumpdir, f"{utt_id}_to_{target_spk}-local.npy"
-                        ),
-                        l_.astype(np.float32),
-                        allow_pickle=False,
-                    )
-            else:
-                raise ValueError("support only hdf5 or npy format.")
-
-
-if __name__ == "__main__":
-    main()

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/path.sh

@@ -1,33 +0,0 @@
-# cuda related
-export CUDA_HOME=/usr/local/cuda-11.3
-export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
-
-# path related
-export PRJ_ROOT="${PWD}/../../.."
-if [ -e "${PRJ_ROOT}/tools/venv/bin/activate" ]; then
-    # shellcheck disable=SC1090
-    . "${PRJ_ROOT}/tools/venv/bin/activate"
-fi
-
-# python related
-export OMP_NUM_THREADS=1
-export PYTHONIOENCODING=UTF-8
-export MPL_BACKEND=Agg
-
-# check installation
-if ! command -v parallel-wavegan-train > /dev/null; then
-    echo "Error: It seems setup is not finished." >&2
-    echo "Error: Please setup your environment by following README.md" >&2
-    return 1
-fi
-if ! command -v jq > /dev/null; then
-    echo "Error: It seems jq is not installed." >&2
-    echo "Error: Please install via \`sudo apt-get install jq\`." >&2
-    echo "Error: If you do not have sudo, please download from https://stedolan.github.io/jq/download/." >&2
-    return 1
-fi
-if ! command -v yq > /dev/null; then
-    echo "Error: It seems yq is not installed." >&2
-    echo "Error: Please install via \`pip install yq\`." >&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/vctk/vq1/run.sh

@@ -1,295 +0,0 @@
-#!/bin/bash
-
-# Copyright 2020 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-. ./cmd.sh || exit 1;
-. ./path.sh || exit 1;
-
-# basic settings
-stage=-1       # stage to start
-stop_stage=100 # stage to stop
-verbose=1      # verbosity level (lower is less info)
-n_gpus=1       # number of gpus in training
-n_jobs=16      # number of parallel jobs in feature extraction
-
-# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
-conf=conf/conditioned_melgan_vae.v3.yaml
-
-# speaker setting
-spks="all" # all or you can choose speakers e.g., "p225 p226 p227 ..."
-
-# directory path setting
-download_dir=downloads # directory to save database
-dumpdir=dump           # directory to dump features
-
-# training related setting
-tag=""     # tag for directory to save model
-resume=""  # checkpoint path to resume training
-           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
-
-# decoding related setting
-checkpoint="" # checkpoint path to be used for decoding
-              # if not provided, the latest one will be used
-              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-train_set="train_nodev_$(echo "${spks}" | tr " " "_")" # name of training data directory
-dev_set="dev_$(echo "${spks}" | tr " " "_")"           # name of development data directory
-eval_set="eval_$(echo "${spks}" | tr " " "_")"         # name of evaluation data directory
-
-set -euo pipefail
-
-if [ "${stage}" -le -1 ] && [ "${stop_stage}" -ge -1 ]; then
-    echo "Stage -1: Data download"
-    local/data_download.sh "${download_dir}"
-fi
-
-if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
-    echo "Stage 0: Data preparation"
-    train_data_dirs=""
-    dev_data_dirs=""
-    eval_data_dirs=""
-    # if set to "all", use all of the speakers in the corpus
-    if [ "${spks}" = "all" ]; then
-        # NOTE(kan-bayashi): p315 will not be used since it lacks txt data
-        spks=$(find "${download_dir}/VCTK-Corpus/wav48" \
-            -maxdepth 1 -name "p*" -exec basename {} \; | sort | grep -v p315)
-    fi
-    for spk in ${spks}; do
-        local/data_prep.sh \
-            --fs "$(yq ".sampling_rate" "${conf}")" \
-            --train_set "train_nodev_${spk}" \
-            --dev_set "dev_${spk}" \
-            --eval_set "eval_${spk}" \
-            "${download_dir}/VCTK-Corpus" "${spk}" data
-        train_data_dirs+=" data/train_nodev_${spk}"
-        dev_data_dirs+=" data/dev_${spk}"
-        eval_data_dirs+=" data/eval_${spk}"
-    done
-    # shellcheck disable=SC2086
-    utils/combine_data.sh "data/${train_set}" ${train_data_dirs}
-    # shellcheck disable=SC2086
-    utils/combine_data.sh "data/${dev_set}" ${dev_data_dirs}
-    # shellcheck disable=SC2086
-    utils/combine_data.sh "data/${eval_set}" ${eval_data_dirs}
-
-    # make spk2idx
-    cut -f 2 -d " " "data/${train_set}/utt2spk" | sort | uniq | \
-        awk '{print $1 " " NR -1}' > "data/${train_set}/spk2idx"
-fi
-
-stats_ext=$(grep -q "hdf5" <(yq ".format" "${conf}") && echo "h5" || echo "npy")
-if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
-    echo "Stage 1: Feature extraction"
-    # extract raw features
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
-        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
-        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
-        if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-            opts="--extract-f0 --skip-mel-ext"
-        else
-            opts="--skip-mel-ext"
-        fi
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
-            parallel-wavegan-preprocess \
-                --config "${conf}" \
-                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
-                --segments "${dumpdir}/${name}/raw/segments.JOB" \
-                --utt2spk "${dumpdir}/${name}/raw/utt2spk.JOB" \
-                --spk2idx "data/${train_set}/spk2idx" \
-                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --verbose "${verbose}" \
-                ${opts}
-        echo "Successfully finished feature extraction of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished feature extraction."
-
-    if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-        # calculate statistics for normalization
-        echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics.log."
-        ${train_cmd} "${dumpdir}/${train_set}/compute_statistics.log" \
-            parallel-wavegan-compute-statistics \
-                --config "${conf}" \
-                --rootdir "${dumpdir}/${train_set}/raw" \
-                --dumpdir "${dumpdir}/${train_set}" \
-                --target-feats "local" \
-                --verbose "${verbose}"
-        echo "Successfully finished calculation of statistics."
-
-        # normalize and dump them
-        pids=()
-        for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-        (
-            [ ! -e "${dumpdir}/${name}/norm" ] && mkdir -p "${dumpdir}/${name}/norm"
-            echo "Nomalization start. See the progress via ${dumpdir}/${name}/norm/normalize.*.log."
-            ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/norm/normalize.JOB.log" \
-                parallel-wavegan-normalize \
-                    --config "${conf}" \
-                    --stats "${dumpdir}/${train_set}/stats.${stats_ext}" \
-                    --rootdir "${dumpdir}/${name}/raw/dump.JOB" \
-                    --dumpdir "${dumpdir}/${name}/norm/dump.JOB" \
-                    --target-feats "local" \
-                    --verbose "${verbose}"
-            echo "Successfully finished normalization of ${name} set."
-        ) &
-        pids+=($!)
-        done
-        i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-        [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-        echo "Successfully finished normalization."
-    fi
-fi
-
-if [ -z "${tag}" ]; then
-    expdir="exp/${train_set}_vctk_$(basename "${conf}" .yaml)"
-else
-    expdir="exp/${train_set}_vctk_${tag}"
-fi
-if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
-    echo "Stage 2: Network training"
-    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
-    if [ "${n_gpus}" -gt 1 ]; then
-        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
-    else
-        train="parallel-wavegan-train"
-    fi
-    if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-        cp "${dumpdir}/${train_set}/stats.${stats_ext}" "${expdir}"
-        train_dumpdir="${dumpdir}/${train_set}/norm"
-        dev_dumpdir="${dumpdir}/${dev_set}/norm"
-    else
-        train_dumpdir="${dumpdir}/${train_set}/raw"
-        dev_dumpdir="${dumpdir}/${dev_set}/raw"
-    fi
-    echo "Training start. See the progress via ${expdir}/train.log."
-    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
-        ${train} \
-            --config "${conf}" \
-            --train-dumpdir "${train_dumpdir}" \
-            --dev-dumpdir "${dev_dumpdir}" \
-            --outdir "${expdir}" \
-            --resume "${resume}" \
-            --verbose "${verbose}"
-    echo "Successfully finished training."
-fi
-
-if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
-    echo "Stage 3: Network decoding"
-    # shellcheck disable=SC2012
-    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
-        # shellcheck disable=SC2030,SC2031
-        [ "${n_gpus}" -gt 1 ] && n_gpus=1
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
-        if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-            decode_dumpdir="${dumpdir}/${name}/norm"
-        else
-            decode_dumpdir="${dumpdir}/${name}/raw"
-        fi
-        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
-            parallel-wavegan-decode \
-                --dumpdir "${decode_dumpdir}" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished decoding."
-fi
-
-if [ "${stage}" -le 4 ] && [ "${stop_stage}" -ge 4 ]; then
-    echo "Stage 4: Voice conversion preparation"
-    if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-        # calculate each speaker statistics for f0 transformation
-        echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics_for_vc.log."
-        ${train_cmd} "${dumpdir}/${train_set}/compute_statistics_for_vc.log" \
-            parallel-wavegan-compute-statistics \
-                --config "${conf}" \
-                --rootdir "${dumpdir}/${train_set}/norm" \
-                --dumpdir "${dumpdir}/${train_set}" \
-                --target-feats "local" \
-                --utt2spk "data/${train_set}/utt2spk" \
-                --verbose "${verbose}"
-        echo "Successfully finished calculation of statistics."
-    fi
-
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-            vc_rootdir="${dumpdir}/${name}/norm"
-            vc_dumpdir="${dumpdir}/${name}/norm_vc"
-        else
-            vc_rootdir="${dumpdir}/${name}/raw"
-            vc_dumpdir="${dumpdir}/${name}/raw_vc"
-        fi
-        echo "Setup for voice conversion start. See the progress via ${vc_dumpdir}/setup_for_vc.*.log."
-        [ ! -e "${vc_dumpdir}" ] && mkdir -p "${vc_dumpdir}"
-        ${train_cmd} JOB=1:${n_jobs} "${vc_dumpdir}/setup_for_vc.JOB.log" \
-            local/setup_for_vc.py \
-                --config "${conf}" \
-                --rootdir "${vc_rootdir}/dump.JOB" \
-                --dumpdir "${vc_dumpdir}/dump.JOB" \
-                --statdir "${dumpdir}/${train_set}" \
-                --utt2spk "data/${name}/utt2spk" \
-                --spk2idx "data/${train_set}/spk2idx" \
-                --verbose "${verbose}"
-        echo "Successfully finished steup of ${name} set for voice conversion."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished setup for voice conversion."
-fi
-
-if [ "${stage}" -le 5 ] && [ "${stop_stage}" -ge 5 ]; then
-    echo "Stage 5: Network decoding for voice conversion"
-    # shellcheck disable=SC2012
-    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
-        # shellcheck disable=SC2030,SC2031
-        [ "${n_gpus}" -gt 1 ] && n_gpus=1
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode_vc.log."
-        if [ "$(yq ".use_local_condition" "${conf}")" = "true" ]; then
-            vc_decode_dumpdir="${dumpdir}/${name}/norm_vc"
-        else
-            vc_decode_dumpdir="${dumpdir}/${name}/raw_vc"
-        fi
-        ${cuda_cmd} JOB=1:${n_jobs} --max-jobs-run 1 --gpu "${n_gpus}" "${outdir}/${name}/decode_vc.JOB.log" \
-            parallel-wavegan-decode \
-                --dumpdir "${vc_decode_dumpdir}/dump.JOB" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}/wav.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set for voice conversion."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished decoding for voice conversion."
-fi
-echo "Finished."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/cmd.sh

@@ -1,91 +0,0 @@
-# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
-# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
-# e.g.
-#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
-#
-# Options:
-#   --time <time>: Limit the maximum time to execute.
-#   --mem <mem>: Limit the maximum memory usage.
-#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
-#   --num-threads <ngpu>: Specify the number of CPU core.
-#   --gpu <ngpu>: Specify the number of GPU devices.
-#   --config: Change the configuration file from default.
-#
-# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
-# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
-# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
-# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
-#
-# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
-# These options are mapping to specific options for each backend and
-# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
-# If jobs failed, your configuration might be wrong for your environment.
-#
-#
-# The official documentaion for run.pl, queue.pl, slurm.pl, and ssh.pl:
-#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
-# =========================================================~
-
-
-# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
-cmd_backend="stdout"
-
-# Local machine, without any Job scheduling system
-if [ "${cmd_backend}" = local ]; then
-
-    # The other usage
-    export train_cmd="utils/run.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/run.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/run.pl"
-
-# Local machine, without any Job scheduling system
-elif [ "${cmd_backend}" = stdout ]; then
-
-    # The other usage
-    export train_cmd="utils/stdout.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/stdout.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/stdout.pl"
-
-# "qsub" (SGE, Torque, PBS, etc.)
-elif [ "${cmd_backend}" = sge ]; then
-    # The default setting is written in conf/queue.conf.
-    # You must change "-q g.q" for the "queue" for your environment.
-    # To know the "queue" names, type "qhost -q"
-    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
-
-    export train_cmd="utils/queue.pl"
-    export cuda_cmd="utils/queue.pl"
-    export decode_cmd="utils/queue.pl"
-
-# "sbatch" (Slurm)
-elif [ "${cmd_backend}" = slurm ]; then
-    # The default setting is written in conf/slurm.conf.
-    # You must change "-p cpu" and "-p gpu" for the "partion" for your environment.
-    # To know the "partion" names, type "sinfo".
-    # You can use "--gpu * " by defualt for slurm and it is interpreted as "--gres gpu:*"
-    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
-
-    export train_cmd="utils/slurm.pl"
-    export cuda_cmd="utils/slurm.pl"
-    export decode_cmd="utils/slurm.pl"
-
-elif [ "${cmd_backend}" = ssh ]; then
-    # You have to create ".queue/machines" to specify the host to execute jobs.
-    # e.g. .queue/machines
-    #   host1
-    #   host2
-    #   host3
-    # Assuming you can login them without any password, i.e. You have to set ssh keys.
-
-    export train_cmd="utils/ssh.pl"
-    export cuda_cmd="utils/ssh.pl"
-    export decode_cmd="utils/ssh.pl"
-
-else
-    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/hifigan.v1.debug.yaml

@@ -1,152 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 3800               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: true       # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: HiFiGANGenerator
-generator_params:
-    in_channels: 80                       # Number of input channels.
-    out_channels: 1                       # Number of output channels.
-    channels: 32                          # Number of initial channels.
-    kernel_size: 7                        # Kernel size of initial and final conv layers.
-    upsample_scales: [8, 8, 2, 2]         # Upsampling scales.
-    upsample_kernel_sizes: [16, 16, 4, 4] # Kernel size for upsampling layers.
-    resblock_kernel_sizes: [3, 7, 11]     # Kernel size for residual blocks.
-    resblock_dilations:                   # Dilations for residual blocks.
-        - [1, 3, 5]
-        - [1, 3, 5]
-        - [1, 3, 5]
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
-    nonlinear_activation: "LeakyReLU"     # Nonlinear activation type.
-    nonlinear_activation_params:          # Nonlinear activation paramters.
-        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
-discriminator_params:
-    scales: 2                             # Number of multi-scale discriminator.
-    scale_downsample_pooling: "AvgPool1d" # Pooling operation for scale discriminator.
-    scale_downsample_pooling_params:
-        kernel_size: 4                    # Pooling kernel size.
-        stride: 2                         # Pooling stride.
-        padding: 2                        # Padding size.
-    scale_discriminator_params:
-        in_channels: 1                    # Number of input channels.
-        out_channels: 1                   # Number of output channels.
-        kernel_sizes: [15, 41, 5, 3]      # List of kernel sizes.
-        channels: 16                      # Initial number of channels.
-        max_downsample_channels: 32       # Maximum number of channels in downsampling conv layers.
-        max_groups: 16                    # Maximum number of groups in downsampling conv layers.
-        bias: true
-        downsample_scales: [4, 4, 4, 4]   # Downsampling scales.
-        nonlinear_activation: "LeakyReLU" # Nonlinear activation.
-        nonlinear_activation_params:
-            negative_slope: 0.1
-    follow_official_norm: true            # Whether to follow the official norm setting.
-    periods: [2, 3]                       # List of period for multi-period discriminator.
-    period_discriminator_params:
-        in_channels: 1                    # Number of input channels.
-        out_channels: 1                   # Number of output channels.
-        kernel_sizes: [5, 3]              # List of kernel sizes.
-        channels: 32                      # Initial number of channels.
-        downsample_scales: [4, 4, 4, 4]   # Downsampling scales.
-        max_downsample_channels: 32       # Maximum number of channels in downsampling conv layers.
-        bias: true                        # Whether to use bias parameter in conv layer."
-        nonlinear_activation: "LeakyReLU" # Nonlinear activation.
-        nonlinear_activation_params:      # Nonlinear activation paramters.
-            negative_slope: 0.1
-        use_weight_norm: true             # Whether to apply weight normalization.
-        use_spectral_norm: false          # Whether to apply spectral normalization.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-use_stft_loss: false
-use_mel_loss: true
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-use_feat_match_loss: true
-feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by # discriminators.
-    average_by_layers: false         # Whether to average loss by # layers in each discriminator.
-    include_final_outputs: true      # Whether to include final outputs in feat match loss calculation.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_aux: 45.0       # Loss balancing coefficient for STFT loss.
-lambda_adv: 1.0        # Loss balancing coefficient for adversarial loss.
-lambda_feat_match: 2.0 # Loss balancing coefficient for feat match loss..
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: AdamW
-generator_optimizer_params:
-    lr: 0.0002            # Generator's learning rate.
-    betas: [0.8, 0.99]    # Generator's optimizer beta
-    weight_decay: 0.0     # Generator's weight decay coefficient.
-generator_scheduler_type: ExponentialLR
-generator_scheduler_params:
-    gamma: 0.999           # Generator's scheduler gamma.
-generator_grad_norm: -1    # Generator's gradient norm.
-discriminator_optimizer_type: AdamW
-discriminator_optimizer_params:
-    lr: 0.0002            # Discriminator's learning rate.
-    betas: [0.8, 0.99]    # Discriminator's optimizer beta
-    weight_decay: 0.0     # Discriminator's weight decay coefficient.
-discriminator_scheduler_type: ExponentialLR
-discriminator_scheduler_params:
-    gamma: 0.999            # Discriminator's scheduler gamma.
-discriminator_grad_norm: -1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-generator_train_start_steps: 5     # Number of steps to start to train discriminator.
-discriminator_train_start_steps: 0 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/melgan.v1.debug.yaml

@@ -1,112 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 4000               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: true       # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "MelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 80               # Number of input channels.
-    out_channels: 1               # Number of output channels.
-    kernel_size: 7                # Kernel size of initial and final conv layers.
-    channels: 256                 # Initial number of channels for conv layers.
-    upsample_scales: [8, 8, 2, 2] # List of Upsampling scales.
-    stack_kernel_size: 3          # Kernel size of dilated conv layers in residual stack.
-    stacks: 1                     # Number of stacks in a single residual stack module.
-    use_weight_norm: True         # Whether to use weight normalization.
-    use_causal_conv: False        # Whether to use causal convolution.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Number of output channels.
-    layers: 10            # Number of conv layers.
-    conv_channels: 32     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    nonlinear_activation: "LeakyReLU" # Nonlinear function after each conv.
-    nonlinear_activation_params:      # Nonlinear function parameters
-        negative_slope: 0.2           # Alpha in LeakyReLU.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 4.0  # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/melgan.v3.debug.yaml

@@ -1,121 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 0                  # Minimum freq in mel basis calculation.
-fmax: 4000               # Maximum frequency in mel basis calculation.
-global_gain_scale: 0.8   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. " npy " or " hdf5 " is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "MelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 80               # Number of input channels.
-    out_channels: 1               # Number of output channels.
-    kernel_size: 7                # Kernel size of initial and final conv layers.
-    channels: 256                 # Initial number of channels for conv layers.
-    upsample_scales: [8, 8, 2, 2] # List of Upsampling scales.
-    stack_kernel_size: 3          # Kernel size of dilated conv layers in residual stack.
-    stacks: 1                     # Number of stacks in a single residual stack module.
-    use_weight_norm: True         # Whether to use weight normalization.
-    use_causal_conv: False        # Whether to use causal convolution.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
-discriminator_params:
-    in_channels: 1                    # Number of input channels.
-    out_channels: 1                   # Number of output channels.
-    scales: 3                         # Number of multi-scales.
-    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
-    downsample_pooling_params:        # Parameters of the above pooling function.
-        kernel_size: 4
-        stride: 2
-        padding: 1
-        count_include_pad: False
-    kernel_sizes: [5, 3]              # List of kernel size.
-    channels: 16                      # Number of channels of the initial conv layer.
-    max_downsample_channels: 64       # Maximum number of channels of downsampling layers.
-    downsample_scales: [4, 4, 4, 4]   # List of downsampling scales.
-    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
-    nonlinear_activation_params:      # Parameters of nonlinear activation function.
-        negative_slope: 0.2
-    use_weight_norm: True             # Whether to use weight norm.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-use_feat_match_loss: true # Whether to use feature matching loss.
-lambda_feat_match: 25.0   # Loss balancing coefficient for feature matching loss.
-lambda_adv: 4.0           # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/multi_band_melgan.v1.debug.yaml

@@ -1,126 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 200            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 0                  # Minimum freq in mel basis calculation.
-fmax: 4000               # Maximum frequency in mel basis calculation.
-global_gain_scale: 0.8   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. " npy " or " hdf5 " is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "MelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 80               # Number of input channels.
-    out_channels: 4               # Number of output channels.
-    kernel_size: 7                # Kernel size of initial and final conv layers.
-    channels: 256                 # Initial number of channels for conv layers.
-    upsample_scales: [2, 5, 5]    # List of Upsampling scales.
-    stack_kernel_size: 3          # Kernel size of dilated conv layers in residual stack.
-    stacks: 1                     # Number of stacks in a single residual stack module.
-    use_weight_norm: True         # Whether to use weight normalization.
-    use_causal_conv: False        # Whether to use causal convolution.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "MelGANMultiScaleDiscriminator" # Discriminator type.
-discriminator_params:
-    in_channels: 1                    # Number of input channels.
-    out_channels: 1                   # Number of output channels.
-    scales: 3                         # Number of multi-scales.
-    downsample_pooling: "AvgPool1d"   # Pooling type for the input downsampling.
-    downsample_pooling_params:        # Parameters of the above pooling function.
-        kernel_size: 4
-        stride: 2
-        padding: 1
-        count_include_pad: False
-    kernel_sizes: [5, 3]              # List of kernel size.
-    channels: 16                      # Number of channels of the initial conv layer.
-    max_downsample_channels: 64       # Maximum number of channels of downsampling layers.
-    downsample_scales: [4, 4, 4]      # List of downsampling scales.
-    nonlinear_activation: "LeakyReLU" # Nonlinear activation function.
-    nonlinear_activation_params:      # Parameters of nonlinear activation function.
-        negative_slope: 0.2
-    use_weight_norm: True             # Whether to use weight norm.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-use_subband_stft_loss: true
-subband_stft_loss_params:
-    fft_sizes: [384, 683, 171]  # List of FFT size for STFT-based loss.
-    hop_sizes: [30, 60, 10]     # List of hop size for STFT-based loss
-    win_lengths: [150, 300, 60] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-use_feat_match_loss: false # Whether to use feature matching loss.
-lambda_adv: 2.5            # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4000       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/parallel_wavegan.v1.debug.diff_fs.yaml

@@ -1,121 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000           # Sampling rate.
-sampling_rate_for_feats: 4000 # Sampling rate for features.
-fft_size: 1024                # FFT size.
-hop_size: 256                 # Hop size.
-win_length: null              # Window length.
-                              # If set to null, it will be the same as fft_size.
-window: "hann"                # Window function.
-num_mels: 80                  # Number of mel basis.
-fmin: 80                      # Minimum freq in mel basis calculation.
-fmax: 2000                    # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0        # Will be multiplied to all of waveform.
-trim_silence: true            # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20      # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024         # Frame size in trimming.
-trim_hop_size: 256            # Hop size in trimming.
-format: "hdf5"                # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Kernel size of dilated convolution.
-    layers: 20            # Number of residual block layers.
-    stacks: 2             # Number of stacks i.e., dilation cycles.
-    residual_channels: 16 # Number of channels in residual conv.
-    gate_channels: 32     # Number of channels in gated conv.
-    skip_channels: 16     # Number of channels in skip conv.
-    aux_channels: 80      # Number of channels for auxiliary feature conv.
-                          # Must be the same as num_mels.
-    aux_context_window: 1 # Context window size for auxiliary feature.
-                          # If set to 2, previous 2 and future 2 frames will be considered.
-    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    upsample_net: "ConvInUpsampleNetwork" # Upsampling network architecture.
-    upsample_params:                      # Upsampling network parameters.
-        upsample_scales: [4, 4, 4, 4]     # Upsampling scales. Prodcut of these must be the same as hop size.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Number of output channels.
-    layers: 10            # Number of conv layers.
-    conv_channels: 16     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    nonlinear_activation: "LeakyReLU" # Nonlinear function after each conv.
-    nonlinear_activation_params:      # Nonlinear function parameters
-        negative_slope: 0.2           # Alpha in LeakyReLU.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 4.0  # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/parallel_wavegan.v1.debug.npy.yaml

@@ -1,121 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 3800               # Maximum frequency in mel basis calculation.
-log_base: null           # Log base.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: true       # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "npy"            # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Kernel size of dilated convolution.
-    layers: 20            # Number of residual block layers.
-    stacks: 2             # Number of stacks i.e., dilation cycles.
-    residual_channels: 16 # Number of channels in residual conv.
-    gate_channels: 32     # Number of channels in gated conv.
-    skip_channels: 16     # Number of channels in skip conv.
-    aux_channels: 80      # Number of channels for auxiliary feature conv.
-                          # Must be the same as num_mels.
-    aux_context_window: 1 # Context window size for auxiliary feature.
-                          # If set to 2, previous 2 and future 2 frames will be considered.
-    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    upsample_net: "ConvInUpsampleNetwork" # Upsampling network architecture.
-    upsample_params:                      # Upsampling network parameters.
-        upsample_scales: [4, 4, 4, 4]     # Upsampling scales. Prodcut of these must be the same as hop size.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Number of output channels.
-    layers: 10            # Number of conv layers.
-    conv_channels: 16     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    nonlinear_activation: "LeakyReLU" # Nonlinear function after each conv.
-    nonlinear_activation_params:      # Nonlinear function parameters
-        negative_slope: 0.2           # Alpha in LeakyReLU.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 4.0  # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/parallel_wavegan.v1.debug.yaml

@@ -1,120 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 3800               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: true       # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Kernel size of dilated convolution.
-    layers: 20            # Number of residual block layers.
-    stacks: 2             # Number of stacks i.e., dilation cycles.
-    residual_channels: 16 # Number of channels in residual conv.
-    gate_channels: 32     # Number of channels in gated conv.
-    skip_channels: 16     # Number of channels in skip conv.
-    aux_channels: 80      # Number of channels for auxiliary feature conv.
-                          # Must be the same as num_mels.
-    aux_context_window: 1 # Context window size for auxiliary feature.
-                          # If set to 2, previous 2 and future 2 frames will be considered.
-    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    upsample_net: "ConvInUpsampleNetwork" # Upsampling network architecture.
-    upsample_params:                      # Upsampling network parameters.
-        upsample_scales: [4, 4, 4, 4]     # Upsampling scales. Prodcut of these must be the same as hop size.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_params:
-    in_channels: 1        # Number of input channels.
-    out_channels: 1       # Number of output channels.
-    kernel_size: 3        # Number of output channels.
-    layers: 10            # Number of conv layers.
-    conv_channels: 16     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
-                          # If set to true, it will be applied to all of the conv layers.
-    nonlinear_activation: "LeakyReLU" # Nonlinear function after each conv.
-    nonlinear_activation_params:      # Nonlinear function parameters
-        negative_slope: 0.2           # Alpha in LeakyReLU.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 4.0  # Loss balancing coefficient.
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_params:
-    lr: 0.0001             # Generator's learning rate.
-    eps: 1.0e-6            # Generator's epsilon.
-    weight_decay: 0.0      # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000      # Generator's scheduler step size.
-    gamma: 0.5             # Generator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: 10    # Generator's gradient norm.
-discriminator_optimizer_params:
-    lr: 0.00005            # Discriminator's learning rate.
-    eps: 1.0e-6            # Discriminator's epsilon.
-    weight_decay: 0.0      # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000      # Discriminator's scheduler step size.
-    gamma: 0.5             # Discriminator's scheduler gamma.
-                           # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: 1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/slurm.conf

@@ -1,12 +0,0 @@
-# Default configuration
-command sbatch --export=PATH  --ntasks-per-node=1
-option time=* --time $0
-option mem=* --mem-per-cpu $0
-option mem=0          # Do not add anything to qsub_opts
-option num_threads=* --cpus-per-task $0 --ntasks-per-node=1
-option num_threads=1 --cpus-per-task 1  --ntasks-per-node=1 # Do not add anything to qsub_opts
-default gpu=0
-option gpu=0 -p cpu
-option gpu=* -p gpu --gres=gpu:$0
-# note: the --max-jobs-run option is supported as a special case
-# by slurm.pl and you don't have to handle it in the config file.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/style_melgan.v1.debug.yaml

@@ -1,135 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 0                  # Minimum freq in mel basis calculation.
-fmax: 4000               # Maximum frequency in mel basis calculation.
-global_gain_scale: 0.8   # Will be multiplied to all of waveform.
-trim_silence: false      # Whether to trim the start and end of silence.
-trim_threshold_in_db: 60 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. " npy " or " hdf5 " is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: "StyleMelGANGenerator" # Generator type.
-generator_params:
-    in_channels: 128
-    aux_channels: 80
-    channels: 64
-    out_channels: 1
-    kernel_size: 9
-    dilation: 2
-    bias: True
-    noise_upsample_scales: [4, 2, 2, 2]
-    noise_upsample_activation: "LeakyReLU"
-    noise_upsample_activation_params:
-        negative_slope: 0.2
-    upsample_scales: [2, 2, 2, 2, 2, 2, 2, 2, 1]
-    upsample_mode: "nearest"
-    gated_function: "softmax"
-    use_weight_norm: True
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: "StyleMelGANDiscriminator" # Discriminator type.
-discriminator_params:
-    repeats: 4
-    window_sizes: [512, 1024, 2048, 4096]
-    pqmf_params:
-        - [1, None, None, None]
-        - [2, 62, 0.26700, 9.0]
-        - [4, 62, 0.14200, 9.0]
-        - [8, 62, 0.07949, 9.0]
-    discriminator_params:
-        out_channels: 1
-        kernel_sizes: [5, 3]
-        channels: 16
-        max_downsample_channels: 512
-        bias: True
-        downsample_scales: [4, 4, 4, 1]
-        nonlinear_activation: "LeakyReLU"
-        nonlinear_activation_params:
-            negative_slope: 0.2
-    use_weight_norm: True
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-stft_loss_params:
-    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
-    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
-    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
-    window: "hann_window"         # Window function for STFT-based loss
-lambda_aux: 1.0                   # Loss balancing coefficient for aux loss.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_adv: 1.0 # Loss balancing coefficient for adv loss.
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-    loss_type: hinge                 # Loss function type ("mse" or "hinge").
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-    loss_type: hinge                 # Loss function type ("mse" or "hinge").
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 8192       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: Adam
-generator_optimizer_params:
-    lr: 0.0001              # Generator's learning rate.
-    betas: [0.5, 0.9]
-    weight_decay: 0.0       # Generator's weight decay coefficient.
-generator_scheduler_params:
-    step_size: 200000       # Generator's scheduler step size.
-    gamma: 0.5              # Generator's scheduler gamma.
-                            # At each step size, lr will be multiplied by this parameter.
-generator_grad_norm: -1     # Generator's gradient norm.
-discriminator_optimizer_type: Adam
-discriminator_optimizer_params:
-    lr: 0.00005             # Discriminator's learning rate.
-    betas: [0.5, 0.9]
-    weight_decay: 0.0       # Discriminator's weight decay coefficient.
-discriminator_scheduler_params:
-    step_size: 200000       # Discriminator's scheduler step size.
-    gamma: 0.5              # Discriminator's scheduler gamma.
-                            # At each step size, lr will be multiplied by this parameter.
-discriminator_grad_norm: -1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-discriminator_train_start_steps: 5 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/conf/uhifigan.v1.debug.yaml

@@ -1,155 +0,0 @@
-# This is the configuration file for yesno dataset.
-# Note that this configuration is just for debugging.
-
-###########################################################
-#                FEATURE EXTRACTION SETTING               #
-###########################################################
-sampling_rate: 8000      # Sampling rate.
-fft_size: 1024           # FFT size.
-hop_size: 256            # Hop size.
-win_length: null         # Window length.
-                         # If set to null, it will be the same as fft_size.
-window: "hann"           # Window function.
-num_mels: 80             # Number of mel basis.
-fmin: 80                 # Minimum freq in mel basis calculation.
-fmax: 3800               # Maximum frequency in mel basis calculation.
-global_gain_scale: 1.0   # Will be multiplied to all of waveform.
-trim_silence: true       # Whether to trim the start and end of silence.
-trim_threshold_in_db: 20 # Need to tune carefully if the recording is not good.
-trim_frame_size: 1024    # Frame size in trimming.
-trim_hop_size: 256       # Hop size in trimming.
-format: "hdf5"           # Feature file format. "npy" or "hdf5" is supported.
-
-###########################################################
-#         GENERATOR NETWORK ARCHITECTURE SETTING          #
-###########################################################
-generator_type: UHiFiGANGenerator
-generator_params:
-    in_channels: 80                       # Number of input channels.
-    out_channels: 1                       # Number of output channels.
-    channels: 32                          # Number of initial channels.
-    kernel_size: 7                        # Kernel size of initial and final conv layers.
-    downsample_scales: [2, 2, 8, 8]       # Downsampling scales.
-    downsample_kernel_sizes: [4, 4, 16, 16] # Kernel size for downsampling layers.
-    upsample_scales: [8, 8, 2, 2]         # Upsampling scales.
-    upsample_kernel_sizes: [16, 16, 4, 4] # Kernel size for upsampling layers.
-    resblock_kernel_sizes: [3, 7, 11]     # Kernel size for residual blocks.
-    resblock_dilations:                   # Dilations for residual blocks.
-        - [1, 3, 5]
-        - [1, 3, 5]
-        - [1, 3, 5]
-    dropout: 0.1                          # Probability of dropout.
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
-    nonlinear_activation: "LeakyReLU"     # Nonlinear activation type.
-    nonlinear_activation_params:          # Nonlinear activation paramters.
-        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
-
-###########################################################
-#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
-###########################################################
-discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
-discriminator_params:
-    scales: 2                             # Number of multi-scale discriminator.
-    scale_downsample_pooling: "AvgPool1d" # Pooling operation for scale discriminator.
-    scale_downsample_pooling_params:
-        kernel_size: 4                    # Pooling kernel size.
-        stride: 2                         # Pooling stride.
-        padding: 2                        # Padding size.
-    scale_discriminator_params:
-        in_channels: 1                    # Number of input channels.
-        out_channels: 1                   # Number of output channels.
-        kernel_sizes: [15, 41, 5, 3]      # List of kernel sizes.
-        channels: 16                      # Initial number of channels.
-        max_downsample_channels: 32       # Maximum number of channels in downsampling conv layers.
-        max_groups: 16                    # Maximum number of groups in downsampling conv layers.
-        bias: true
-        downsample_scales: [4, 4, 4, 4]   # Downsampling scales.
-        nonlinear_activation: "LeakyReLU" # Nonlinear activation.
-        nonlinear_activation_params:
-            negative_slope: 0.1
-    follow_official_norm: true            # Whether to follow the official norm setting.
-    periods: [2, 3]                       # List of period for multi-period discriminator.
-    period_discriminator_params:
-        in_channels: 1                    # Number of input channels.
-        out_channels: 1                   # Number of output channels.
-        kernel_sizes: [5, 3]              # List of kernel sizes.
-        channels: 32                      # Initial number of channels.
-        downsample_scales: [4, 4, 4, 4]   # Downsampling scales.
-        max_downsample_channels: 32       # Maximum number of channels in downsampling conv layers.
-        bias: true                        # Whether to use bias parameter in conv layer."
-        nonlinear_activation: "LeakyReLU" # Nonlinear activation.
-        nonlinear_activation_params:      # Nonlinear activation paramters.
-            negative_slope: 0.1
-        use_weight_norm: true             # Whether to apply weight normalization.
-        use_spectral_norm: false          # Whether to apply spectral normalization.
-
-###########################################################
-#                   STFT LOSS SETTING                     #
-###########################################################
-use_stft_loss: false
-use_mel_loss: true
-generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
-use_feat_match_loss: true
-feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by # discriminators.
-    average_by_layers: false         # Whether to average loss by # layers in each discriminator.
-    include_final_outputs: true      # Whether to include final outputs in feat match loss calculation.
-
-###########################################################
-#               ADVERSARIAL LOSS SETTING                  #
-###########################################################
-lambda_aux: 45.0       # Loss balancing coefficient for STFT loss.
-lambda_adv: 1.0        # Loss balancing coefficient for adversarial loss.
-lambda_feat_match: 2.0 # Loss balancing coefficient for feat match loss..
-
-###########################################################
-#                  DATA LOADER SETTING                    #
-###########################################################
-batch_size: 2               # Batch size.
-batch_max_steps: 4096       # Length of each audio in batch. Make sure dividable by hop_size.
-pin_memory: true            # Whether to pin memory in Pytorch DataLoader.
-num_workers: 2              # Number of workers in Pytorch DataLoader.
-remove_short_samples: false # Whether to remove samples the length of which are less than batch_max_steps.
-allow_cache: true           # Whether to allow cache in dataset. If true, it requires cpu memory.
-
-###########################################################
-#             OPTIMIZER & SCHEDULER SETTING               #
-###########################################################
-generator_optimizer_type: AdamW
-generator_optimizer_params:
-    lr: 0.0002            # Generator's learning rate.
-    betas: [0.8, 0.99]    # Generator's optimizer beta
-    weight_decay: 0.0     # Generator's weight decay coefficient.
-generator_scheduler_type: ExponentialLR
-generator_scheduler_params:
-    gamma: 0.999           # Generator's scheduler gamma.
-generator_grad_norm: -1    # Generator's gradient norm.
-discriminator_optimizer_type: AdamW
-discriminator_optimizer_params:
-    lr: 0.0002            # Discriminator's learning rate.
-    betas: [0.8, 0.99]    # Discriminator's optimizer beta
-    weight_decay: 0.0     # Discriminator's weight decay coefficient.
-discriminator_scheduler_type: ExponentialLR
-discriminator_scheduler_params:
-    gamma: 0.999            # Discriminator's scheduler gamma.
-discriminator_grad_norm: -1 # Discriminator's gradient norm.
-
-###########################################################
-#                    INTERVAL SETTING                     #
-###########################################################
-generator_train_start_steps: 5     # Number of steps to start to train discriminator.
-discriminator_train_start_steps: 0 # Number of steps to start to train discriminator.
-train_max_steps: 10                # Number of training steps.
-save_interval_steps: 5             # Interval steps to save checkpoint.
-eval_interval_steps: 5             # Interval steps to evaluate the network.
-log_interval_steps: 5              # Interval steps to record the training log.
-
-###########################################################
-#                     OTHER SETTING                       #
-###########################################################
-num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/local/data_download.sh

@@ -1,28 +0,0 @@
-#!/bin/bash
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-download_dir=$1
-
-# check arguments
-if [ $# != 1 ]; then
-    echo "Usage: $0 <download_dir>"
-    exit 1
-fi
-
-set -euo pipefail
-
-cwd=$(pwd)
-if [ ! -e "${download_dir}/LJSpeech-1.1" ]; then
-    mkdir -p "${download_dir}"
-    cd "${download_dir}"
-    wget http://www.openslr.org/resources/1/waves_yesno.tar.gz
-    tar -xvzf waves_yesno.tar.gz
-    rm ./waves_yesno/README*
-    rm waves_yesno.tar.gz
-    cd "${cwd}"
-    echo "Successfully downloaded data."
-else
-    echo "Already exists. Skipped."
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/local/data_prep.sh

@@ -1,88 +0,0 @@
-#!/bin/bash
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-# shellcheck disable=SC1091
-. ./path.sh || exit 1;
-
-num_dev=10
-num_eval=10
-train_set="train_nodev"
-dev_set="dev"
-eval_set="eval"
-shuffle=false
-use_fake_segments=false
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-db_root=$1
-data_dir=$2
-
-# check arguments
-if [ $# != 2 ]; then
-    echo "Usage: $0 [Options] <db_root> <data_dir>"
-    echo "e.g.: $0 downloads/waves_yesno data"
-    echo ""
-    echo "Options:"
-    echo "    --num_dev: number of development uttreances (default=10)."
-    echo "    --num_eval: number of evaluation uttreances (default=10)."
-    echo "    --train_set: name of train set (default=train_nodev)."
-    echo "    --dev_set: name of dev set (default=dev)."
-    echo "    --eval_set: name of eval set (default=eval)."
-    echo "    --shuffle: whether to perform shuffle in making dev / eval set (default=false)."
-    echo "    --use_fake_segments: whether to use fake segments (default=false)."
-    exit 1
-fi
-
-set -euo pipefail
-
-[ ! -e "${data_dir}/all" ] && mkdir -p "${data_dir}/all"
-
-# set filenames
-scp="${data_dir}/all/wav.scp"
-segments="${data_dir}/all/segments"
-
-# check file existence
-[ -e "${scp}" ] && rm "${scp}"
-[ -e "${segments}" ] && rm "${segments}"
-
-# make all scp
-find "${db_root}" -follow -name "*.wav" | sort | while read -r filename; do
-    id=$(basename "${filename}" | sed -e "s/\.[^\.]*$//g")
-    echo "${id} ${filename}" >> "${scp}"
-    # NOTE(kan-bayashi): for integration test
-    if "${use_fake_segments}"; then
-        echo "${id}_1 ${id} 0.0 $(soxi -D "${filename}")" >> "${data_dir}/all/segments"
-        echo "${id}_2 ${id} 0.0 $(soxi -D "${filename}")" >> "${data_dir}/all/segments"
-    fi
-done
-
-# split
-num_all=$(wc -l < "${scp}")
-if "${use_fake_segments}"; then
-    num_all=$(wc -l < "${segments}")
-fi
-num_deveval=$((num_dev + num_eval))
-num_train=$((num_all - num_deveval))
-utils/split_data.sh \
-    --num_first "${num_train}" \
-    --num_second "${num_deveval}" \
-    --shuffle "${shuffle}" \
-    "${data_dir}/all" \
-    "${data_dir}/${train_set}" \
-    "${data_dir}/deveval"
-utils/split_data.sh \
-    --num_first "${num_dev}" \
-    --num_second "${num_eval}" \
-    --shuffle "${shuffle}" \
-    "${data_dir}/deveval" \
-    "${data_dir}/${dev_set}" \
-    "${data_dir}/${eval_set}"
-
-# remove tmp directories
-rm -rf "${data_dir}/all"
-rm -rf "${data_dir}/deveval"
-
-echo "Successfully prepared data."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/path.sh

@@ -1,33 +0,0 @@
-# cuda related
-export CUDA_HOME=/usr/local/cuda-11.3
-export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
-
-# path related
-export PRJ_ROOT="${PWD}/../../.."
-if [ -e "${PRJ_ROOT}/tools/venv/bin/activate" ]; then
-    # shellcheck disable=SC1090
-    . "${PRJ_ROOT}/tools/venv/bin/activate"
-fi
-
-# python related
-export OMP_NUM_THREADS=1
-export PYTHONIOENCODING=UTF-8
-export MPL_BACKEND=Agg
-
-# check installation
-if ! command -v parallel-wavegan-train > /dev/null; then
-    echo "Error: It seems setup is not finished." >&2
-    echo "Error: Please setup your environment by following README.md" >&2
-    return 1
-fi
-if ! command -v jq > /dev/null; then
-    echo "Error: It seems jq is not installed." >&2
-    echo "Error: Please install via \`sudo apt-get install jq\`." >&2
-    echo "Error: If you do not have sudo, please download from https://stedolan.github.io/jq/download/." >&2
-    return 1
-fi
-if ! command -v yq > /dev/null; then
-    echo "Error: It seems yq is not installed." >&2
-    echo "Error: Please install via \`pip install yq\`." >&2
-    return 1
-fi

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/voc1/run.sh

@@ -1,178 +0,0 @@
-#!/bin/bash
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-. ./cmd.sh || exit 1;
-. ./path.sh || exit 1;
-
-# basic settings
-stage=-1       # stage to start
-stop_stage=100 # stage to stop
-verbose=1      # verbosity level (lower is less info)
-n_gpus=0       # number of gpus in training
-n_jobs=2       # number of parallel jobs in feature extraction
-
-# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
-conf=conf/parallel_wavegan.v1.debug.yaml
-
-# directory path setting
-download_dir=downloads # direcotry to save downloaded files
-dumpdir=dump           # directory to dump features
-
-# data setting
-use_fake_segments=false  # for testing
-
-# training related setting
-tag=""     # tag for directory to save model
-resume=""  # checkpoint path to resume training
-           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
-
-# decoding related setting
-checkpoint="" # checkpoint path to be used for decoding
-              # if not provided, the latest one will be used
-              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)
-
-# shellcheck disable=SC1091
-. utils/parse_options.sh || exit 1;
-
-train_set="train_nodev" # name of training data directory
-dev_set="dev"           # name of development data direcotry
-eval_set="eval"         # name of evaluation data direcotry
-
-set -euo pipefail
-
-if [ "${stage}" -le -1 ] && [ "${stop_stage}" -ge -1 ]; then
-    echo "Stage -1: Data download"
-    local/data_download.sh "${download_dir}"
-fi
-
-if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
-    echo "Stage 0: Data preparation"
-    local/data_prep.sh \
-        --use_fake_segments "${use_fake_segments}" \
-        --train_set "${train_set}" \
-        --dev_set "${dev_set}" \
-        --eval_set "${eval_set}" \
-        "${download_dir}/waves_yesno" data
-fi
-
-stats_ext=$(grep -q "hdf5" <(yq ".format" "${conf}") && echo "h5" || echo "npy")
-if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
-    echo "Stage 1: Feature extraction"
-    # extract raw features
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
-        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
-        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
-            parallel-wavegan-preprocess \
-                --config "${conf}" \
-                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
-                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished feature extraction of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished feature extraction."
-
-    # calculate statistics for normalization
-    echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics.log."
-    ${train_cmd} "${dumpdir}/${train_set}/compute_statistics.log" \
-        parallel-wavegan-compute-statistics \
-            --config "${conf}" \
-            --rootdir "${dumpdir}/${train_set}/raw" \
-            --dumpdir "${dumpdir}/${train_set}" \
-            --verbose "${verbose}"
-    echo "Successfully finished calculation of statistics."
-
-    # normalize and dump them
-    pids=()
-    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${dumpdir}/${name}/norm" ] && mkdir -p "${dumpdir}/${name}/norm"
-        echo "Nomalization start. See the progress via ${dumpdir}/${name}/norm/normalize.*.log."
-        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/norm/normalize.JOB.log" \
-            parallel-wavegan-normalize \
-                --config "${conf}" \
-                --stats "${dumpdir}/${train_set}/stats.${stats_ext}" \
-                --rootdir "${dumpdir}/${name}/raw/dump.JOB" \
-                --dumpdir "${dumpdir}/${name}/norm/dump.JOB" \
-                --verbose "${verbose}"
-        echo "Successfully finished normalization of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished normalization."
-fi
-
-if [ -z "${tag}" ]; then
-    expdir="exp/${train_set}_yesno_$(basename "${conf}" .yaml)"
-else
-    expdir="exp/${train_set}_yesno_${tag}"
-fi
-if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
-    echo "Stage 2: Network training"
-    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
-    cp "${dumpdir}/${train_set}/stats.${stats_ext}" "${expdir}"
-    if [ "${n_gpus}" -gt 1 ]; then
-        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
-    else
-        train="parallel-wavegan-train"
-    fi
-    echo "Training start. See the progress via ${expdir}/train.log."
-    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
-        ${train} \
-            --config "${conf}" \
-            --train-dumpdir "${dumpdir}/${train_set}/norm" \
-            --dev-dumpdir "${dumpdir}/${dev_set}/norm" \
-            --outdir "${expdir}" \
-            --resume "${resume}" \
-            --verbose "${verbose}"
-    echo "Successfully finished training."
-fi
-
-if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
-    echo "Stage 3: Network decoding"
-    # shellcheck disable=SC2012
-    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
-    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
-    pids=()
-    for name in "${dev_set}" "${eval_set}"; do
-    (
-        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
-        [ "${n_gpus}" -gt 1 ] && n_gpus=1
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
-        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
-            parallel-wavegan-decode \
-                --dumpdir "${dumpdir}/${name}/norm" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set."
-
-        # NOTE(kan-bayashi): Extra decoding for debugging
-        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
-        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
-            parallel-wavegan-decode \
-                --normalize-before \
-                --dumpdir "${dumpdir}/${name}/raw" \
-                --checkpoint "${checkpoint}" \
-                --outdir "${outdir}/${name}" \
-                --verbose "${verbose}"
-        echo "Successfully finished decoding of ${name} set."
-    ) &
-    pids+=($!)
-    done
-    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
-    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
-    echo "Successfully finished decoding."
-fi
-echo "Finished."

audio/shiftyspeech/speech_synthesis/vocoders/ParallelWaveGAN/egs/yesno/vq1/cmd.sh

@@ -1,91 +0,0 @@
-# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
-# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
-# e.g.
-#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
-#
-# Options:
-#   --time <time>: Limit the maximum time to execute.
-#   --mem <mem>: Limit the maximum memory usage.
-#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
-#   --num-threads <ngpu>: Specify the number of CPU core.
-#   --gpu <ngpu>: Specify the number of GPU devices.
-#   --config: Change the configuration file from default.
-#
-# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
-# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
-# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
-# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
-#
-# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
-# These options are mapping to specific options for each backend and
-# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
-# If jobs failed, your configuration might be wrong for your environment.
-#
-#
-# The official documentaion for run.pl, queue.pl, slurm.pl, and ssh.pl:
-#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
-# =========================================================~
-
-
-# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
-cmd_backend="stdout"
-
-# Local machine, without any Job scheduling system
-if [ "${cmd_backend}" = local ]; then
-
-    # The other usage
-    export train_cmd="utils/run.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/run.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/run.pl"
-
-# Local machine, without any Job scheduling system
-elif [ "${cmd_backend}" = stdout ]; then
-
-    # The other usage
-    export train_cmd="utils/stdout.pl"
-    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
-    export cuda_cmd="utils/stdout.pl"
-    # Used for "*_recog.py"
-    export decode_cmd="utils/stdout.pl"
-
-# "qsub" (SGE, Torque, PBS, etc.)
-elif [ "${cmd_backend}" = sge ]; then
-    # The default setting is written in conf/queue.conf.
-    # You must change "-q g.q" for the "queue" for your environment.
-    # To know the "queue" names, type "qhost -q"
-    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
-
-    export train_cmd="utils/queue.pl"
-    export cuda_cmd="utils/queue.pl"
-    export decode_cmd="utils/queue.pl"
-
-# "sbatch" (Slurm)
-elif [ "${cmd_backend}" = slurm ]; then
-    # The default setting is written in conf/slurm.conf.
-    # You must change "-p cpu" and "-p gpu" for the "partion" for your environment.
-    # To know the "partion" names, type "sinfo".
-    # You can use "--gpu * " by defualt for slurm and it is interpreted as "--gres gpu:*"
-    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
-
-    export train_cmd="utils/slurm.pl"
-    export cuda_cmd="utils/slurm.pl"
-    export decode_cmd="utils/slurm.pl"
-
-elif [ "${cmd_backend}" = ssh ]; then
-    # You have to create ".queue/machines" to specify the host to execute jobs.
-    # e.g. .queue/machines
-    #   host1
-    #   host2
-    #   host3
-    # Assuming you can login them without any password, i.e. You have to set ssh keys.
-
-    export train_cmd="utils/ssh.pl"
-    export cuda_cmd="utils/ssh.pl"
-    export decode_cmd="utils/ssh.pl"
-
-else
-    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
-    return 1
-fi