Spaces:
Running on Zero
Running on Zero
File size: 4,496 Bytes
4a9b62b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 | #!/usr/bin/env python3
# -*- coding: utf-8 -*-
# Copyright 2019 Tomoki Hayashi
# MIT License (https://opensource.org/licenses/MIT)
# MIT License
# Copyright (c) 2025 Yiwei Guo
# This file is based on ParallelWaveGAN (https://github.com/kan-bayashi/ParallelWaveGAN).
# Above is the original copyright notice.
"""Encode with Trained LSCodec Encoder."""
import argparse
import logging
import os
import time
import numpy as np
import re
import librosa
import soundfile as sf
import torch
from pathlib import Path
import yaml
from tqdm import tqdm
import kaldiio
from lscodec.utils import load_model, load_audio
from kaldiio import WriteHelper
def main():
"""Run decoding process."""
parser = argparse.ArgumentParser(
description="Encode features with trained LSCodec"
)
parser.add_argument(
"--wav-scp",
"--scp",
default=None,
type=str,
help="kaldi-style wav.scp file. "
"you need to specify wav-scp.",
)
parser.add_argument(
"--outdir",
type=str,
required=True,
help="directory to save generated speech.",
)
parser.add_argument(
"--pretrained-dir",
type=str,
default="pretrained/",
help="directory to save pretrained model. (default=None). It will need to contain lscodec_encoder.pt and encoder_config.yml",
)
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
config_path = os.path.join(args.pretrained_dir, "encoder_config.yml")
checkpoint_path = os.path.join(args.pretrained_dir, "lscodec_encoder.pt")
with open(config_path) as f:
config = yaml.load(f, Loader=yaml.Loader)
config.update(vars(args))
config['pretrain_codebook'] = os.path.join(args.pretrained_dir, "codebook.npy") # overwrite
# check arguments
if args.wav_scp is None:
raise ValueError("Please specify --wav-scp.")
utt2path = dict()
with open(args.wav_scp, 'r') as fr:
for line in fr.readlines():
terms = line.strip().split()
utt2path[terms[0]] = terms[1]
logging.info(f"The number of features to be decoded = {len(utt2path)}.")
# setup model
if torch.cuda.is_available():
device = torch.device("cuda")
logging.info("Using GPU.")
else:
device = torch.device("cpu")
logging.info("Using CPU.")
model = load_model(config, checkpoint_path)
logging.info(f"Loaded model parameters from {checkpoint_path}.")
model = model.eval().to(device)
# start generation
outdir = str(Path(args.outdir).absolute())
with torch.no_grad(), tqdm(utt2path.items(), desc="[encode]") as pbar, WriteHelper(f"ark,scp:{outdir}/feats.ark,{outdir}/feats.scp") as writer:
for i, (utt_id, audio_path) in enumerate(pbar, 1):
audio, sr = load_audio(audio_path)
if sr != config['feature_extractor_params']['fs']:
audio_for_extractor = librosa.resample(audio, orig_sr=sr, target_sr=config['feature_extractor_params']['fs'])
else:
audio_for_extractor = audio
audio_for_extractor = torch.tensor(audio_for_extractor).float().to(device) # (L)
audio_for_extractor = audio_for_extractor.unsqueeze(0).unsqueeze(0) # (1, 1, L)
means, _, embed_index = model.encode(audio_for_extractor)
# now the embed_index is [L, 1] token tensor
if embed_index is not None:
writer(utt_id, embed_index.float().cpu().numpy())
else:
writer(utt_id, means.squeeze(0).cpu().numpy())
if __name__ == "__main__":
main()
|