Update v0.2.0 models, reference-free 2.4 kb/s codec, C baselines, PESQ-WB/STOI metrics

README.md

@@ -8,8 +8,7 @@ tags:
 - speech-coding
 - neural-audio-codec
 - wavenet
-- codec2
-- speaker-verification
+- arithmetic-coding
 - icassp2018
 - safetensors
 - gguf
@@ -19,7 +18,7 @@ datasets:
 - vctk
 ---
 
-# WaveNet Closed-Loop Waveform Speech Coder (WW, ICASSP 2018)
+# WaveNet Closed-Loop Waveform Speech Coder (`WW`, ICASSP 2018)
 
 **This is not an officially supported Google product.**
 
@@ -31,83 +30,65 @@ datasets:
 
 ## Model Summary
 
-Unconditioned closed-loop autoregressive WaveNet speech coder (Section 2.2 & Section 3.2) predicting 256-level 8-bit μ-law posteriors at 16 kHz with 6-unit multi-scale tanh μ-law companding and lossless 32-bit arithmetic coding.
+Unconditioned closed-loop autoregressive WaveNet speech coder (`WW`, Section 2.3 & Section 3.2) predicting 256-level 8-bit μ-law categorical posteriors at 16 kHz (`2` blocks × `10` dilated causal residual layers) paired with a bit-exact 32-bit integer arithmetic coder.
 
 ### Exported Artifacts Included in This Repository
 
 1. `model.safetensors`: Hugging Face SafeTensors FP32 weights.
-2. `config.json`: Complete model architecture and hyperparameter specification.
+2. `config.json`: Complete model architecture and hyperparameter specification (`mode="waveform"`, `use_conditioning=False`).
 3. `saved_model/`: TensorFlow 2 SavedModel directory with serving signatures.
-4. `model_fp32.tflite` & `model_int8.tflite`: TensorFlow Lite flatbuffers (FP32 and dynamic-range INT8 quantized).
-5. `model_f16.gguf` & `model_q4_k_m.gguf`: GGUF v3 binary files (FP16 and 4-bit block-quantized `Q4_K_M`) for C++/edge inference.
+4. `model_fp32.tflite` (`685,348` bytes) & `model_int8.tflite` (`181,248` bytes): Verified TensorFlow Lite flatbuffers.
+5. `model_f16.gguf` & `model_q4_k_m.gguf`: Verified GGUF v3 binary files (`164` tensors) for C++/edge inference.
 6. `training_metrics.json` & `evaluation_metrics.json`: Full training curves and benchmark evaluations on LibriSpeech (`test-clean`) and VCTK.
 
-## Training Progression
+## Training Progression (`training_metrics.json`)
 
-- **Initial Eval Cross-Entropy**: `5.4979` bits/sample
-- **Final Eval Cross-Entropy**: `5.2410` bits/sample
-- **Training Windows**: `48` (`187.79 s`)
+- **Training Corpus**: `160` speaker-disjoint utterances (`64` windows) from LibriSpeech `test-clean` + CSTR VCTK at 16 kHz (`-18.4 dBFS` active level)
+- **Wall-Clock Training Time**: `14.85 s` (`10` epochs)
+- **Initial Eval Cross-Entropy**: `7.0186` bits/sample (`4.8649` nats/sample)
+- **Final Eval Cross-Entropy**: `4.0603` bits/sample (`2.8144` nats/sample)
 
-## Evaluation Results
+## Evaluation Results — Closed-Loop Waveform Coding Rates (Section 3.2)
 
-### 1. Closed-Loop Waveform Coding Rate & Entropy (Section 3.2)
+| Dataset / Split | Mean Conditional Entropy H̄ (bits/sample) | Theoretical Lower-Bound Bitrate (kb/s) | Cross-Entropy Rate R (bits/sample) | Cross-Entropy Bitrate (kb/s) | Active Arithmetic Coded Rate (bits/sample) | Active Arithmetic Bitrate (kb/s) | Arithmetic Coder Lossless Match Rate |
+| :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| **Paper Reported (WSJ0, Sec. 3.2, 60-layer conditioned WW)** | 2.6113 (1.81 nats) | 41.78 | 2.6546 (1.84 nats) | 42.47 (~42 kb/s) | — | — | — |
+| **Reproduced — LibriSpeech (`test-clean`, 16 utt)** | **4.8556** | **77.69** | **5.2570** (3.6438 nats) | **84.11** | **5.0156** | **80.25** | **100.0% (16/16)** |
+| **Reproduced — VCTK Corpus (16 utt)** | **4.8433** | **77.49** | **5.0888** (3.5273 nats) | **81.42** | **4.8990** | **78.38** | **100.0% (16/16)** |
 
-| Dataset | Mean Conditional Entropy H̄ (bits/sample) | Mean Entropy Rate (kb/s) | Cross-Entropy Rate R (bits/sample) | Cross-Entropy Bitrate (kb/s) | Arithmetic Coder Lossless Match Rate |
-| :--- | :---: | :---: | :---: | :---: | :---: |
-| **LibriSpeech (`test-clean`)** | 5.7415 | 91.86 | 5.9333 | 94.93 | 100.0% |
-| **VCTK Corpus** | 5.7457 | 91.93 | 5.8275 | 93.24 | 100.0% |
+### Waveform Quality Comparison (`WW` vs. 8-bit μ-law & 16 kHz Reference)
 
-### 2. Parametric 2.4 kb/s & Reference Codec Comparison — LibriSpeech (`test-clean`)
-
-| System / Codec | Bitrate (kb/s) | Objective Wideband MOS-LQO (1–5) | MCD (dB) ↓ | LSD (dB) ↓ | SegSNR (dB) ↑ | High-Band (4–8 kHz) Energy (%) |
-| :--- | :---: | :---: | :---: | :---: | :---: | :---: |
-| **Reference (16 kHz)** | 256.00 | 4.850 | 0.000 | 0.000 | 45.000 | 3.641% |
-| **8-bit mu-law (128 kb/s)** | 128.00 | 4.738 | 0.305 | 0.641 | 37.484 | 3.650% |
-| **WaveNet Waveform (WW)** | 94.93 | 4.738 | 0.305 | 0.641 | 37.484 | 3.650% |
-| **AMR-WB (23.85 kb/s)** | 23.85 | 3.449 | 3.197 | 5.307 | 8.900 | 3.397% |
-| **WaveNet Parametric (W_w, 2.4 kb/s)** | 2.40 | 3.245 | 3.420 | 3.565 | 3.716 | 3.671% |
-| **WaveNet Parametric (W_wo, 2.4 kb/s)** | 2.40 | 2.230 | 7.505 | 7.905 | 1.537 | 1.643% |
-| **MELP (2.4 kb/s)** | 2.40 | 1.627 | 8.412 | 8.770 | -2.744 | 0.041% |
-| **Codec 2 (2.4 kb/s)** | 2.40 | 1.646 | 8.277 | 8.172 | -2.491 | 0.023% |
-| **Speex (2.15 kb/s)** | 2.15 | 1.893 | 7.459 | 8.219 | 0.426 | 0.069% |
-
-### 3. Parametric 2.4 kb/s & Reference Codec Comparison — VCTK Corpus
-
-| System / Codec | Bitrate (kb/s) | Objective Wideband MOS-LQO (1–5) | MCD (dB) ↓ | LSD (dB) ↓ | SegSNR (dB) ↑ | High-Band (4–8 kHz) Energy (%) |
-| :--- | :---: | :---: | :---: | :---: | :---: | :---: |
-| **Reference (16 kHz)** | 256.00 | 4.850 | 0.000 | 0.000 | 45.000 | 1.826% |
-| **8-bit mu-law (128 kb/s)** | 128.00 | 4.697 | 0.501 | 0.648 | 37.162 | 1.834% |
-| **WaveNet Waveform (WW)** | 93.24 | 4.697 | 0.501 | 0.648 | 37.162 | 1.834% |
-| **AMR-WB (23.85 kb/s)** | 23.85 | 3.364 | 3.541 | 5.359 | 8.749 | 1.563% |
-| **WaveNet Parametric (W_w, 2.4 kb/s)** | 2.40 | 3.110 | 4.210 | 4.056 | 4.352 | 2.064% |
-| **WaveNet Parametric (W_wo, 2.4 kb/s)** | 2.40 | 2.131 | 8.193 | 8.397 | 1.972 | 1.510% |
-| **MELP (2.4 kb/s)** | 2.40 | 1.454 | 9.397 | 9.090 | -2.683 | 0.033% |
-| **Codec 2 (2.4 kb/s)** | 2.40 | 1.356 | 10.187 | 8.785 | -2.545 | 0.015% |
-| **Speex (2.15 kb/s)** | 2.15 | 1.722 | 8.392 | 8.341 | 0.257 | 0.038% |
-
-### 4. Text-Independent Speaker Verification & Triangle Test (Section 3.4)
-
-| Dataset | 8-bit μ-law EER (%) [95% CI] | WaveNet W_wo (2.4 kb/s) EER (%) [95% CI] | WaveNet W_w (2.4 kb/s) EER (%) [95% CI] | Triangle Test Accuracy (W_w closer to μ-law than W_wo) |
-| :--- | :---: | :---: | :---: | :---: |
-| **LibriSpeech (`test-clean`)** | 12.05% [2.22%, 19.64%] | 25.00% [9.75%, 37.96%] | 25.00% [5.36%, 50.00%] | 43.25% (Δcos = +0.3206) |
-| **VCTK Corpus** | 25.00% [11.60%, 42.44%] | 37.05% [20.04%, 50.52%] | 24.55% [12.05%, 46.89%] | 42.60% (Δcos = +0.2990) |
+| Dataset | Codec | Bitrate (kb/s) | ITU-T P.862.2 PESQ-WB ↑ | STOI (0–1) ↑ | F0 RMSE (Hz) ↓ | MCD (dB) ↓ | LSD (dB) ↓ | SegSNR (dB) ↑ | High-Band (4–8 kHz) Energy (%) |
+| :--- | :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| **LibriSpeech (`test-clean`)** | **Reference (16 kHz)** | 256.00 | 4.644 | 1.000 | 0.000 | 0.000 | 0.000 | 45.000 | 4.062% |
+| **LibriSpeech (`test-clean`)** | **8-bit μ-law (128 kb/s)** | 128.00 | 4.171 | 0.999 | 0.184 | 0.427 | 0.827 | 35.420 | 4.072% |
+| **LibriSpeech (`test-clean`)** | **WaveNet Waveform (`WW`)** | **84.11 (80.25 active)** | **4.171** | **0.999** | **0.184** | **0.427** | **0.827** | **35.420** | **4.072%** |
+| **VCTK Corpus** | **Reference (16 kHz)** | 256.00 | 4.644 | 1.000 | 0.000 | 0.000 | 0.000 | 45.000 | 1.774% |
+| **VCTK Corpus** | **8-bit μ-law (128 kb/s)** | 128.00 | 3.772 | 0.999 | 0.107 | 0.472 | 0.810 | 34.976 | 1.783% |
+| **VCTK Corpus** | **WaveNet Waveform (`WW`)** | **81.42 (78.38 active)** | **3.772** | **0.999** | **0.107** | **0.472** | **0.810** | **34.976** | **1.783%** |
 
 ## Usage Examples
 
 ### Load SafeTensors in Python (`wavenet-coding`)
 
 ```python
+import numpy as np
 from huggingface_hub import snapshot_download
 from wavenet_coding import inference
 
 model_dir = snapshot_download("wq2012/wavenet-waveform-coder")
-codec = inference.WaveNetParametricCodec.from_pretrained(model_dir)
+codec = inference.WaveNetWaveformCodec.from_pretrained(model_dir)
+
+t = np.linspace(0.0, 0.2, 3200, endpoint=False, dtype=np.float32)
+waveform_16k = 0.12 * np.sin(2.0 * np.pi * 180.0 * t)
+result = codec.encode_and_decode(waveform_16k, run_arithmetic_coder=True)
+print("Lossless exact match:", result["lossless_exact_match"])
+print("Active arithmetic rate (kb/s):", result["arithmetic_kbps_active"])
 ```
 
 ### Run TFLite Flatbuffer (`model_int8.tflite`)
 
 ```python
-import numpy as np
 import tensorflow as tf
 
 interpreter = tf.lite.Interpreter(model_path="model_int8.tflite")

evaluation_metrics.json

@@ -1,324 +1,436 @@
 {
-  "model_id": "wq2012/wavenet-waveform-coder",
-  "training_metrics": {
-    "task": "wavenet",
-    "mode": "waveform",
-    "speaker_overlapping": false,
-    "num_train_windows": 48,
-    "num_eval_windows": 16,
-    "training_time_sec": 187.79,
-    "initial_eval_bits_per_sample": 5.497945785522461,
-    "final_eval_bits_per_sample": 5.241023063659668,
-    "history": [
-      {
-        "epoch": 1,
-        "train_loss_bits_per_sample": 6.706890265146892,
-        "eval_cross_entropy_bits_per_sample": 5.695452690124512
-      },
-      {
-        "epoch": 2,
-        "train_loss_bits_per_sample": 6.386214256286621,
-        "eval_cross_entropy_bits_per_sample": 5.431291580200195
-      },
-      {
-        "epoch": 3,
-        "train_loss_bits_per_sample": 6.12939437230428,
-        "eval_cross_entropy_bits_per_sample": 5.330618858337402
-      },
-      {
-        "epoch": 4,
-        "train_loss_bits_per_sample": 5.888708432515462,
-        "eval_cross_entropy_bits_per_sample": 5.3388352394104
-      },
-      {
-        "epoch": 5,
-        "train_loss_bits_per_sample": 5.701914151509603,
-        "eval_cross_entropy_bits_per_sample": 5.285624980926514
-      },
-      {
-        "epoch": 6,
-        "train_loss_bits_per_sample": 5.6265692710876465,
-        "eval_cross_entropy_bits_per_sample": 5.294033050537109
-      },
-      {
-        "epoch": 7,
-        "train_loss_bits_per_sample": 5.596172014872233,
-        "eval_cross_entropy_bits_per_sample": 5.288949012756348
-      },
-      {
-        "epoch": 8,
-        "train_loss_bits_per_sample": 5.530875523885091,
-        "eval_cross_entropy_bits_per_sample": 5.241023063659668
-      }
-    ],
-    "exported_files": {
-      "config_path": "pretrained_models/wavenet_waveform_coder/config.json",
-      "safetensors_path": "pretrained_models/wavenet_waveform_coder/model.safetensors",
-      "saved_model_dir": "pretrained_models/wavenet_waveform_coder/saved_model",
-      "tflite": {
-        "fp32_path": "pretrained_models/wavenet_waveform_coder/model_fp32.tflite",
-        "fp32_size_bytes": 2697196,
-        "fp32_verified": true,
-        "int8_path": "pretrained_models/wavenet_waveform_coder/model_int8.tflite",
-        "int8_size_bytes": 924192,
-        "int8_verified": true,
-        "output_shape": [
-          1,
-          320,
-          256
-        ]
-      },
-      "gguf_f16_path": "pretrained_models/wavenet_waveform_coder/model_f16.gguf",
-      "gguf_q4_k_m_path": "pretrained_models/wavenet_waveform_coder/model_q4_k_m.gguf",
-      "gguf_verified_tensors": 164
-    }
-  },
-  "librispeech_test_clean_evaluation": {
+  "librispeech_test_clean": {
     "num_eval_utterances": 16,
     "section_3_2_waveform_coding_rates": {
-      "mean_entropy_bits_per_sample": 5.7415,
-      "mean_entropy_kbps": 91.86,
-      "cross_entropy_bits_per_sample": 5.9333,
-      "cross_entropy_rate_kbps": 94.93,
+      "mean_entropy_bits_per_sample": 4.8556,
+      "mean_entropy_kbps": 77.69,
+      "cross_entropy_bits_per_sample": 5.257,
+      "cross_entropy_nll_nats_per_sample": 3.6438,
+      "cross_entropy_rate_kbps": 84.11,
+      "arithmetic_coded_bits_per_sample": 5.0156,
+      "arithmetic_coded_rate_kbps": 80.25,
       "lossless_exact_match_rate": 1.0
     },
     "section_3_3_codec_comparison": {
       "Reference (16 kHz)": {
         "bitrate_kbps": 256.0,
-        "objective_mos_lqo": 4.85,
+        "pesq_wb_mos_lqo": 4.644,
+        "stoi": 1.0,
+        "f0_rmse_hz": 0.0,
+        "objective_mos_lqo": 4.644,
         "mcd_db": 0.0,
         "lsd_db": 0.0,
         "segmental_snr_db": 45.0,
-        "high_band_energy_ratio_percent": 3.641
+        "high_band_energy_ratio_percent": 4.062
       },
       "8-bit mu-law (128 kb/s)": {
         "bitrate_kbps": 128.0,
-        "objective_mos_lqo": 4.738,
-        "mcd_db": 0.305,
-        "lsd_db": 0.641,
-        "segmental_snr_db": 37.484,
-        "high_band_energy_ratio_percent": 3.65
+        "pesq_wb_mos_lqo": 4.171,
+        "stoi": 0.999,
+        "f0_rmse_hz": 0.184,
+        "objective_mos_lqo": 4.171,
+        "mcd_db": 0.427,
+        "lsd_db": 0.827,
+        "segmental_snr_db": 35.42,
+        "high_band_energy_ratio_percent": 4.072
       },
       "WaveNet Waveform (WW)": {
-        "bitrate_kbps": 94.932,
-        "objective_mos_lqo": 4.738,
-        "mcd_db": 0.305,
-        "lsd_db": 0.641,
-        "segmental_snr_db": 37.484,
-        "high_band_energy_ratio_percent": 3.65
-      },
-      "AMR-WB (23.85 kb/s)": {
-        "bitrate_kbps": 23.85,
-        "objective_mos_lqo": 3.449,
-        "mcd_db": 3.197,
-        "lsd_db": 5.307,
-        "segmental_snr_db": 8.9,
-        "high_band_energy_ratio_percent": 3.397
+        "bitrate_kbps": 84.111,
+        "pesq_wb_mos_lqo": 4.171,
+        "stoi": 0.999,
+        "f0_rmse_hz": 0.184,
+        "objective_mos_lqo": 4.171,
+        "mcd_db": 0.427,
+        "lsd_db": 0.827,
+        "segmental_snr_db": 35.42,
+        "high_band_energy_ratio_percent": 4.072
+      },
+      "Hybrid Mode-Switching (WW + W_w)": {
+        "bitrate_kbps": 50.09,
+        "pesq_wb_mos_lqo": 1.871,
+        "stoi": 0.936,
+        "f0_rmse_hz": 17.721,
+        "objective_mos_lqo": 1.871,
+        "mcd_db": 2.586,
+        "lsd_db": 3.036,
+        "segmental_snr_db": 21.533,
+        "high_band_energy_ratio_percent": 3.88
+      },
+      "AMR-WB (23.05 kb/s)": {
+        "bitrate_kbps": 23.05,
+        "pesq_wb_mos_lqo": 3.964,
+        "stoi": 0.984,
+        "f0_rmse_hz": 8.114,
+        "objective_mos_lqo": 3.964,
+        "mcd_db": 1.491,
+        "lsd_db": 3.131,
+        "segmental_snr_db": 6.049,
+        "high_band_energy_ratio_percent": 2.165
       },
       "WaveNet Parametric (W_w, 2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 3.245,
-        "mcd_db": 3.42,
-        "lsd_db": 3.565,
-        "segmental_snr_db": 3.716,
-        "high_band_energy_ratio_percent": 3.671
+        "pesq_wb_mos_lqo": 1.528,
+        "stoi": 0.85,
+        "f0_rmse_hz": 25.42,
+        "objective_mos_lqo": 1.528,
+        "mcd_db": 5.26,
+        "lsd_db": 5.557,
+        "segmental_snr_db": -2.306,
+        "high_band_energy_ratio_percent": 0.706
       },
       "WaveNet Parametric (W_wo, 2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 2.23,
-        "mcd_db": 7.505,
-        "lsd_db": 7.905,
-        "segmental_snr_db": 1.537,
-        "high_band_energy_ratio_percent": 1.643
+        "pesq_wb_mos_lqo": 1.513,
+        "stoi": 0.85,
+        "f0_rmse_hz": 24.588,
+        "objective_mos_lqo": 1.513,
+        "mcd_db": 5.611,
+        "lsd_db": 5.732,
+        "segmental_snr_db": -2.297,
+        "high_band_energy_ratio_percent": 0.614
       },
       "MELP (2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 1.627,
-        "mcd_db": 8.412,
-        "lsd_db": 8.77,
-        "segmental_snr_db": -2.744,
-        "high_band_energy_ratio_percent": 0.041
+        "pesq_wb_mos_lqo": 1.686,
+        "stoi": 0.778,
+        "f0_rmse_hz": 40.092,
+        "objective_mos_lqo": 1.686,
+        "mcd_db": 8.111,
+        "lsd_db": 7.895,
+        "segmental_snr_db": -2.822,
+        "high_band_energy_ratio_percent": 0.02
       },
       "Codec 2 (2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 1.646,
-        "mcd_db": 8.277,
-        "lsd_db": 8.172,
-        "segmental_snr_db": -2.491,
-        "high_band_energy_ratio_percent": 0.023
+        "pesq_wb_mos_lqo": 1.597,
+        "stoi": 0.847,
+        "f0_rmse_hz": 36.549,
+        "objective_mos_lqo": 1.597,
+        "mcd_db": 8.45,
+        "lsd_db": 8.067,
+        "segmental_snr_db": -2.32,
+        "high_band_energy_ratio_percent": 0.005
       },
       "Speex (2.15 kb/s)": {
         "bitrate_kbps": 2.15,
-        "objective_mos_lqo": 1.893,
-        "mcd_db": 7.459,
-        "lsd_db": 8.219,
-        "segmental_snr_db": 0.426,
-        "high_band_energy_ratio_percent": 0.069
+        "pesq_wb_mos_lqo": 1.472,
+        "stoi": 0.789,
+        "f0_rmse_hz": 31.137,
+        "objective_mos_lqo": 1.472,
+        "mcd_db": 7.944,
+        "lsd_db": 7.807,
+        "segmental_snr_db": -1.38,
+        "high_band_energy_ratio_percent": 0.023
       }
     },
     "section_3_4_speaker_identification": {
       "mulaw_speech_eer": {
-        "eer_percent": 12.053571428571429,
-        "eer_ci95_low_percent": 2.2209821428571432,
-        "eer_ci95_high_percent": 19.642857142857142,
-        "ci_95_low_percent": 2.2209821428571432,
-        "ci_95_high_percent": 19.642857142857142,
-        "threshold": 0.38931805399917674,
-        "mean_genuine_cosine": 0.5900887784230047,
-        "mean_impostor_cosine": 0.06415446283955947
+        "eer_percent": 12.5,
+        "eer_ci95_low_percent": 2.6674107142857144,
+        "eer_ci95_high_percent": 21.886160714285715,
+        "ci_95_low_percent": 2.6674107142857144,
+        "ci_95_high_percent": 21.886160714285715,
+        "threshold": 0.29185709569381924,
+        "mean_genuine_cosine": 0.44530730656223655,
+        "mean_impostor_cosine": -0.031165612260486617
       },
-      "wavenet_parametric_w_wo_eer": {
+      "wavenet_parametric_w_w_eer": {
         "eer_percent": 25.0,
-        "eer_ci95_low_percent": 9.754464285714286,
-        "eer_ci95_high_percent": 37.957589285714285,
-        "ci_95_low_percent": 9.754464285714286,
-        "ci_95_high_percent": 37.957589285714285,
-        "threshold": 0.8321953156106268,
-        "mean_genuine_cosine": 0.8887528008846554,
-        "mean_impostor_cosine": 0.7299079903992016
+        "eer_ci95_low_percent": 13.392857142857142,
+        "eer_ci95_high_percent": 38.404017857142854,
+        "ci_95_low_percent": 13.392857142857142,
+        "ci_95_high_percent": 38.404017857142854,
+        "threshold": 0.3927898672726676,
+        "mean_genuine_cosine": 0.49871190258880344,
+        "mean_impostor_cosine": 0.19032626539991265
       },
-      "wavenet_parametric_w_w_eer": {
+      "wavenet_parametric_w_wo_eer_mulaw_verifier": {
         "eer_percent": 25.0,
-        "eer_ci95_low_percent": 5.357142857142857,
-        "eer_ci95_high_percent": 50.0,
-        "ci_95_low_percent": 5.357142857142857,
-        "ci_95_high_percent": 50.0,
-        "threshold": 0.47203084832451203,
-        "mean_genuine_cosine": 0.616406649899784,
-        "mean_impostor_cosine": 0.22985314070694823
+        "eer_ci95_low_percent": 12.94642857142857,
+        "eer_ci95_high_percent": 41.51785714285714,
+        "ci_95_low_percent": 12.94642857142857,
+        "ci_95_high_percent": 41.51785714285714,
+        "threshold": 0.49189550553056544,
+        "mean_genuine_cosine": 0.5446005744263802,
+        "mean_impostor_cosine": 0.23289187596733665
+      },
+      "wavenet_parametric_w_wo_eer": {
+        "eer_percent": 22.32142857142857,
+        "eer_ci95_low_percent": 8.482142857142858,
+        "eer_ci95_high_percent": 32.611607142857146,
+        "ci_95_low_percent": 8.482142857142858,
+        "ci_95_high_percent": 32.611607142857146,
+        "threshold": 0.21622103255006364,
+        "mean_genuine_cosine": 0.37797472937253285,
+        "mean_impostor_cosine": -0.03277759994961506
+      },
+      "codec2_2400_eer": {
+        "eer_percent": 27.23214285714286,
+        "eer_ci95_low_percent": 17.39955357142857,
+        "eer_ci95_high_percent": 46.886160714285715,
+        "ci_95_low_percent": 17.39955357142857,
+        "ci_95_high_percent": 46.886160714285715,
+        "threshold": 0.6951031882653882,
+        "mean_genuine_cosine": 0.7511831514622034,
+        "mean_impostor_cosine": 0.5752668362534713
+      },
+      "melp_2400_eer": {
+        "eer_percent": 25.44642857142857,
+        "eer_ci95_low_percent": 13.816964285714286,
+        "eer_ci95_high_percent": 46.875,
+        "ci_95_low_percent": 13.816964285714286,
+        "ci_95_high_percent": 46.875,
+        "threshold": 0.6622801368095736,
+        "mean_genuine_cosine": 0.7499099000126639,
+        "mean_impostor_cosine": 0.49901395755586514
+      },
+      "speex_2150_eer": {
+        "eer_percent": 33.48214285714286,
+        "eer_ci95_low_percent": 15.613839285714286,
+        "eer_ci95_high_percent": 41.07142857142857,
+        "ci_95_low_percent": 15.613839285714286,
+        "ci_95_high_percent": 41.07142857142857,
+        "threshold": 0.654569947588562,
+        "mean_genuine_cosine": 0.7099157823552019,
+        "mean_impostor_cosine": 0.5312576204882272
+      },
+      "amr_wb_23050_eer": {
+        "eer_percent": 13.392857142857142,
+        "eer_ci95_low_percent": 3.560267857142857,
+        "eer_ci95_high_percent": 23.22544642857143,
+        "ci_95_low_percent": 3.560267857142857,
+        "ci_95_high_percent": 23.22544642857143,
+        "threshold": 0.33419504474263173,
+        "mean_genuine_cosine": 0.5394971779051225,
+        "mean_impostor_cosine": 0.0811744304899468
       },
       "triangle_test_w_w_vs_w_wo": {
-        "mean_cosine_orig_vs_w_wo": 0.48838181445399975,
-        "mean_cosine_orig_vs_w_w": 0.8089489559047142,
-        "mean_cosine_w_wo_vs_w_w": 0.7808174503733418,
-        "mean_cosine_advantage": 0.3205671414507144,
-        "triangle_test_identification_percent": 43.24508910582853,
-        "triangle_accuracy_percent": 43.24508910582853,
+        "mean_cosine_orig_vs_w_wo": 0.5684091566528269,
+        "mean_cosine_orig_vs_w_w": 0.5917583187780218,
+        "mean_cosine_w_wo_vs_w_w": 0.3938628149217174,
+        "mean_cosine_advantage": 0.02334916212519502,
+        "triangle_test_identification_percent": 50.0,
+        "triangle_accuracy_percent": 50.0,
+        "abx_preference_percent": 68.75,
+        "abx_preference_w_w_over_w_wo_percent": 68.75,
         "random_chance_baseline_percent": 33.33
       }
     }
   },
-  "vctk_evaluation": {
+  "vctk": {
     "num_eval_utterances": 16,
     "section_3_2_waveform_coding_rates": {
-      "mean_entropy_bits_per_sample": 5.7457,
-      "mean_entropy_kbps": 91.93,
-      "cross_entropy_bits_per_sample": 5.8275,
-      "cross_entropy_rate_kbps": 93.24,
+      "mean_entropy_bits_per_sample": 4.8433,
+      "mean_entropy_kbps": 77.49,
+      "cross_entropy_bits_per_sample": 5.0888,
+      "cross_entropy_nll_nats_per_sample": 3.5273,
+      "cross_entropy_rate_kbps": 81.42,
+      "arithmetic_coded_bits_per_sample": 4.899,
+      "arithmetic_coded_rate_kbps": 78.38,
       "lossless_exact_match_rate": 1.0
     },
     "section_3_3_codec_comparison": {
       "Reference (16 kHz)": {
         "bitrate_kbps": 256.0,
-        "objective_mos_lqo": 4.85,
+        "pesq_wb_mos_lqo": 4.644,
+        "stoi": 1.0,
+        "f0_rmse_hz": 0.0,
+        "objective_mos_lqo": 4.644,
         "mcd_db": 0.0,
         "lsd_db": 0.0,
         "segmental_snr_db": 45.0,
-        "high_band_energy_ratio_percent": 1.826
+        "high_band_energy_ratio_percent": 1.774
       },
       "8-bit mu-law (128 kb/s)": {
         "bitrate_kbps": 128.0,
-        "objective_mos_lqo": 4.697,
-        "mcd_db": 0.501,
-        "lsd_db": 0.648,
-        "segmental_snr_db": 37.162,
-        "high_band_energy_ratio_percent": 1.834
+        "pesq_wb_mos_lqo": 3.772,
+        "stoi": 0.999,
+        "f0_rmse_hz": 0.107,
+        "objective_mos_lqo": 3.772,
+        "mcd_db": 0.472,
+        "lsd_db": 0.81,
+        "segmental_snr_db": 34.976,
+        "high_band_energy_ratio_percent": 1.783
       },
       "WaveNet Waveform (WW)": {
-        "bitrate_kbps": 93.24,
-        "objective_mos_lqo": 4.697,
-        "mcd_db": 0.501,
-        "lsd_db": 0.648,
-        "segmental_snr_db": 37.162,
-        "high_band_energy_ratio_percent": 1.834
-      },
-      "AMR-WB (23.85 kb/s)": {
-        "bitrate_kbps": 23.85,
-        "objective_mos_lqo": 3.364,
-        "mcd_db": 3.541,
-        "lsd_db": 5.359,
-        "segmental_snr_db": 8.749,
-        "high_band_energy_ratio_percent": 1.563
+        "bitrate_kbps": 81.421,
+        "pesq_wb_mos_lqo": 3.772,
+        "stoi": 0.999,
+        "f0_rmse_hz": 0.107,
+        "objective_mos_lqo": 3.772,
+        "mcd_db": 0.472,
+        "lsd_db": 0.81,
+        "segmental_snr_db": 34.976,
+        "high_band_energy_ratio_percent": 1.783
+      },
+      "Hybrid Mode-Switching (WW + W_w)": {
+        "bitrate_kbps": 44.382,
+        "pesq_wb_mos_lqo": 1.43,
+        "stoi": 0.938,
+        "f0_rmse_hz": 13.676,
+        "objective_mos_lqo": 1.43,
+        "mcd_db": 3.444,
+        "lsd_db": 3.82,
+        "segmental_snr_db": 19.642,
+        "high_band_energy_ratio_percent": 1.289
+      },
+      "AMR-WB (23.05 kb/s)": {
+        "bitrate_kbps": 23.05,
+        "pesq_wb_mos_lqo": 3.777,
+        "stoi": 0.99,
+        "f0_rmse_hz": 5.78,
+        "objective_mos_lqo": 3.777,
+        "mcd_db": 1.742,
+        "lsd_db": 3.522,
+        "segmental_snr_db": 5.589,
+        "high_band_energy_ratio_percent": 1.113
       },
       "WaveNet Parametric (W_w, 2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 3.11,
-        "mcd_db": 4.21,
-        "lsd_db": 4.056,
-        "segmental_snr_db": 4.352,
-        "high_band_energy_ratio_percent": 2.064
+        "pesq_wb_mos_lqo": 1.345,
+        "stoi": 0.866,
+        "f0_rmse_hz": 33.577,
+        "objective_mos_lqo": 1.345,
+        "mcd_db": 5.792,
+        "lsd_db": 5.988,
+        "segmental_snr_db": -2.276,
+        "high_band_energy_ratio_percent": 0.309
       },
       "WaveNet Parametric (W_wo, 2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 2.131,
-        "mcd_db": 8.193,
-        "lsd_db": 8.397,
-        "segmental_snr_db": 1.972,
-        "high_band_energy_ratio_percent": 1.51
+        "pesq_wb_mos_lqo": 1.368,
+        "stoi": 0.857,
+        "f0_rmse_hz": 36.464,
+        "objective_mos_lqo": 1.368,
+        "mcd_db": 6.107,
+        "lsd_db": 6.131,
+        "segmental_snr_db": -2.223,
+        "high_band_energy_ratio_percent": 0.291
       },
       "MELP (2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 1.454,
-        "mcd_db": 9.397,
-        "lsd_db": 9.09,
-        "segmental_snr_db": -2.683,
-        "high_band_energy_ratio_percent": 0.033
+        "pesq_wb_mos_lqo": 1.486,
+        "stoi": 0.794,
+        "f0_rmse_hz": 33.49,
+        "objective_mos_lqo": 1.486,
+        "mcd_db": 9.908,
+        "lsd_db": 9.171,
+        "segmental_snr_db": -2.74,
+        "high_band_energy_ratio_percent": 0.01
       },
       "Codec 2 (2.4 kb/s)": {
         "bitrate_kbps": 2.4,
-        "objective_mos_lqo": 1.356,
-        "mcd_db": 10.187,
-        "lsd_db": 8.785,
-        "segmental_snr_db": -2.545,
-        "high_band_energy_ratio_percent": 0.015
+        "pesq_wb_mos_lqo": 1.427,
+        "stoi": 0.839,
+        "f0_rmse_hz": 43.198,
+        "objective_mos_lqo": 1.427,
+        "mcd_db": 9.775,
+        "lsd_db": 9.24,
+        "segmental_snr_db": -2.114,
+        "high_band_energy_ratio_percent": 0.002
       },
       "Speex (2.15 kb/s)": {
         "bitrate_kbps": 2.15,
-        "objective_mos_lqo": 1.722,
-        "mcd_db": 8.392,
-        "lsd_db": 8.341,
-        "segmental_snr_db": 0.257,
-        "high_band_energy_ratio_percent": 0.038
+        "pesq_wb_mos_lqo": 1.491,
+        "stoi": 0.82,
+        "f0_rmse_hz": 31.791,
+        "objective_mos_lqo": 1.491,
+        "mcd_db": 9.161,
+        "lsd_db": 8.986,
+        "segmental_snr_db": -1.183,
+        "high_band_energy_ratio_percent": 0.017
       }
     },
     "section_3_4_speaker_identification": {
       "mulaw_speech_eer": {
-        "eer_percent": 25.0,
-        "eer_ci95_low_percent": 11.595982142857144,
-        "eer_ci95_high_percent": 42.44419642857144,
-        "ci_95_low_percent": 11.595982142857144,
-        "ci_95_high_percent": 42.44419642857144,
-        "threshold": 0.2121775659214813,
-        "mean_genuine_cosine": 0.3935697310774072,
-        "mean_impostor_cosine": 0.0036658837067926322
-      },
-      "wavenet_parametric_w_wo_eer": {
-        "eer_percent": 37.05357142857143,
-        "eer_ci95_low_percent": 20.044642857142858,
-        "eer_ci95_high_percent": 50.524553571428584,
-        "ci_95_low_percent": 20.044642857142858,
-        "ci_95_high_percent": 50.524553571428584,
-        "threshold": 0.8427541969049193,
-        "mean_genuine_cosine": 0.8848814072587434,
-        "mean_impostor_cosine": 0.8005241801782267
+        "eer_percent": 12.5,
+        "eer_ci95_low_percent": 3.571428571428571,
+        "eer_ci95_high_percent": 37.511160714285715,
+        "ci_95_low_percent": 3.571428571428571,
+        "ci_95_high_percent": 37.511160714285715,
+        "threshold": 0.40297561601478865,
+        "mean_genuine_cosine": 0.533335601866815,
+        "mean_impostor_cosine": -0.0027530994077539566
       },
       "wavenet_parametric_w_w_eer": {
+        "eer_percent": 21.875,
+        "eer_ci95_low_percent": 8.917410714285715,
+        "eer_ci95_high_percent": 37.94642857142857,
+        "ci_95_low_percent": 8.917410714285715,
+        "ci_95_high_percent": 37.94642857142857,
+        "threshold": 0.4407421890061377,
+        "mean_genuine_cosine": 0.556131588883851,
+        "mean_impostor_cosine": 0.23974738299126758
+      },
+      "wavenet_parametric_w_wo_eer_mulaw_verifier": {
         "eer_percent": 24.55357142857143,
-        "eer_ci95_low_percent": 12.053571428571429,
-        "eer_ci95_high_percent": 46.886160714285715,
-        "ci_95_low_percent": 12.053571428571429,
-        "ci_95_high_percent": 46.886160714285715,
-        "threshold": 0.4774760105853959,
-        "mean_genuine_cosine": 0.6204015199832023,
-        "mean_impostor_cosine": 0.18491605664772884
+        "eer_ci95_low_percent": 11.149553571428573,
+        "eer_ci95_high_percent": 37.94642857142857,
+        "ci_95_low_percent": 11.149553571428573,
+        "ci_95_high_percent": 37.94642857142857,
+        "threshold": 0.4748982079014201,
+        "mean_genuine_cosine": 0.5765287804912442,
+        "mean_impostor_cosine": 0.31020436523949757
+      },
+      "wavenet_parametric_w_wo_eer": {
+        "eer_percent": 15.17857142857143,
+        "eer_ci95_low_percent": 3.571428571428571,
+        "eer_ci95_high_percent": 37.94642857142857,
+        "ci_95_low_percent": 3.571428571428571,
+        "ci_95_high_percent": 37.94642857142857,
+        "threshold": 0.28079196762738023,
+        "mean_genuine_cosine": 0.4166328782518146,
+        "mean_impostor_cosine": -0.01283012844203055
+      },
+      "codec2_2400_eer": {
+        "eer_percent": 25.0,
+        "eer_ci95_low_percent": 5.357142857142857,
+        "eer_ci95_high_percent": 45.2120535714286,
+        "ci_95_low_percent": 5.357142857142857,
+        "ci_95_high_percent": 45.2120535714286,
+        "threshold": 0.7697445190926964,
+        "mean_genuine_cosine": 0.827981640492962,
+        "mean_impostor_cosine": 0.7004077353235066
+      },
+      "melp_2400_eer": {
+        "eer_percent": 25.0,
+        "eer_ci95_low_percent": 11.573660714285715,
+        "eer_ci95_high_percent": 37.94642857142857,
+        "ci_95_low_percent": 11.573660714285715,
+        "ci_95_high_percent": 37.94642857142857,
+        "threshold": 0.792323185831342,
+        "mean_genuine_cosine": 0.8643444983345618,
+        "mean_impostor_cosine": 0.685341582600922
+      },
+      "speex_2150_eer": {
+        "eer_percent": 25.0,
+        "eer_ci95_low_percent": 12.5,
+        "eer_ci95_high_percent": 37.94642857142857,
+        "ci_95_low_percent": 12.5,
+        "ci_95_high_percent": 37.94642857142857,
+        "threshold": 0.7967896929997185,
+        "mean_genuine_cosine": 0.8585711865620879,
+        "mean_impostor_cosine": 0.7166751079752773
+      },
+      "amr_wb_23050_eer": {
+        "eer_percent": 12.5,
+        "eer_ci95_low_percent": 3.560267857142857,
+        "eer_ci95_high_percent": 35.279017857142854,
+        "ci_95_low_percent": 3.560267857142857,
+        "ci_95_high_percent": 35.279017857142854,
+        "threshold": 0.49967441003977797,
+        "mean_genuine_cosine": 0.6482315032144372,
+        "mean_impostor_cosine": 0.17124600824055422
       },
       "triangle_test_w_w_vs_w_wo": {
-        "mean_cosine_orig_vs_w_wo": 0.2806833882483422,
-        "mean_cosine_orig_vs_w_w": 0.5796352731348118,
-        "mean_cosine_w_wo_vs_w_w": 0.8056075167477617,
-        "mean_cosine_advantage": 0.2989518848864696,
-        "triangle_test_identification_percent": 42.59985923688519,
-        "triangle_accuracy_percent": 42.59985923688519,
+        "mean_cosine_orig_vs_w_wo": 0.6013400573986093,
+        "mean_cosine_orig_vs_w_w": 0.6146092914303704,
+        "mean_cosine_w_wo_vs_w_w": 0.413028135873362,
+        "mean_cosine_advantage": 0.013269234031761046,
+        "triangle_test_identification_percent": 43.75,
+        "triangle_accuracy_percent": 43.75,
+        "abx_preference_percent": 56.25,
+        "abx_preference_w_w_over_w_wo_percent": 56.25,
         "random_chance_baseline_percent": 33.33
       }
     }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5ee38e3aef0669a9c1cdfddca0a261bb25ce16114cd4fb5e03d6e771a63d1c32
+oid sha256:5fc80525545624f11741ea635184bb8a85d9c10341733638a2b38a25f85e4e13
 size 2515752

model_f16.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2bfabb7d7bbb5a99c74cedad67114411f0fa6e1ffee9ecd5bf7a0497c018618c
+oid sha256:cbc995bdc362c8c8d7190ae8f5899cfd491c7fc15abf5abf2a4e4dc17476f049
 size 1260672

model_fp32.tflite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:349a157f2d13cb207d5871008cd3a9bee5006dbc691b171639d46fcef58ca2c7
-size 2697196
+oid sha256:88822ca3d5f7611d151d7e65680897f9c67daad93129888f624252987079cc9a
+size 685348

model_int8.tflite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa5b8a273eff56b5c5ebd8fc1cbd05c242cbf626528c129a3c855899bc2e80ed
-size 924192
+oid sha256:cb9cc88b4389d03d5e88ee3484ab66225659b3957470be9024c79938a8c5c4a5
+size 181248

model_q4_k_m.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64794f1066b752ff4e348728be7685d05d623c45cb2ca58c6bbd9ac51b971888
+oid sha256:994bbde1b325d5c728ae74bf197ea7d523a35084e4dae81ba3b5561d6c873060
 size 371584

saved_model/fingerprint.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c98159a31c8e94fd0dd0ea8df01300bf649721f76c0c45ad98bb08946e0940c
+size 100

saved_model/saved_model.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:453c432e2284c30a6a431ead77f8db151a958420de0338ffef7ccd7b0051feee
-size 4376639
+oid sha256:0187b00318130231e736afd01180fe68a4e65d43310fa23958812606d8c760e3
+size 1450963

saved_model/variables/variables.data-00000-of-00001

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f21124fa47b370bd82abea3413760673e8ace57a7be72f044601a1b9ddae62ea
-size 2620208
+oid sha256:3e265c3e92d5f928467ebf0ee1aea840567353ebeaa6932dc3effe55ef9231fa
+size 2560444

saved_model/variables/variables.index

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2716e5100d89623fed95f5161c43f808e37ae3c606dc4642385b18e242fcd9a1
-size 10388
+oid sha256:b7f09afa9f71b397f58e12224e07485233ebd48f5e5b5168a7b7f38757f45a0e
+size 10404

training_metrics.json

@@ -2,51 +2,74 @@
   "task": "wavenet",
   "mode": "waveform",
   "speaker_overlapping": false,
-  "num_train_windows": 48,
+  "num_train_utterances": 160,
+  "num_train_windows": 64,
   "num_eval_windows": 16,
-  "training_time_sec": 187.79,
-  "initial_eval_bits_per_sample": 5.497945785522461,
-  "final_eval_bits_per_sample": 5.241023063659668,
+  "training_time_sec": 14.85,
+  "initial_eval_bits_per_sample": 7.0186,
+  "initial_eval_nll_nats_per_sample": 4.8649,
+  "final_eval_bits_per_sample": 4.0603,
+  "final_eval_nll_nats_per_sample": 2.8144,
   "history": [
     {
       "epoch": 1,
-      "train_loss_bits_per_sample": 6.706890265146892,
-      "eval_cross_entropy_bits_per_sample": 5.695452690124512
+      "train_loss_bits_per_sample": 6.892,
+      "eval_cross_entropy_bits_per_sample": 6.7692,
+      "eval_nll_nats_per_sample": 4.6921
     },
     {
       "epoch": 2,
-      "train_loss_bits_per_sample": 6.386214256286621,
-      "eval_cross_entropy_bits_per_sample": 5.431291580200195
+      "train_loss_bits_per_sample": 6.6984,
+      "eval_cross_entropy_bits_per_sample": 6.5453,
+      "eval_nll_nats_per_sample": 4.5368
     },
     {
       "epoch": 3,
-      "train_loss_bits_per_sample": 6.12939437230428,
-      "eval_cross_entropy_bits_per_sample": 5.330618858337402
+      "train_loss_bits_per_sample": 6.5196,
+      "eval_cross_entropy_bits_per_sample": 6.3332,
+      "eval_nll_nats_per_sample": 4.3898
     },
     {
       "epoch": 4,
-      "train_loss_bits_per_sample": 5.888708432515462,
-      "eval_cross_entropy_bits_per_sample": 5.3388352394104
+      "train_loss_bits_per_sample": 6.3548,
+      "eval_cross_entropy_bits_per_sample": 6.131,
+      "eval_nll_nats_per_sample": 4.2497
     },
     {
       "epoch": 5,
-      "train_loss_bits_per_sample": 5.701914151509603,
-      "eval_cross_entropy_bits_per_sample": 5.285624980926514
+      "train_loss_bits_per_sample": 6.2047,
+      "eval_cross_entropy_bits_per_sample": 5.9357,
+      "eval_nll_nats_per_sample": 4.1143
     },
     {
       "epoch": 6,
-      "train_loss_bits_per_sample": 5.6265692710876465,
-      "eval_cross_entropy_bits_per_sample": 5.294033050537109
+      "train_loss_bits_per_sample": 6.0753,
+      "eval_cross_entropy_bits_per_sample": 5.7441,
+      "eval_nll_nats_per_sample": 3.9815
     },
     {
       "epoch": 7,
-      "train_loss_bits_per_sample": 5.596172014872233,
-      "eval_cross_entropy_bits_per_sample": 5.288949012756348
+      "train_loss_bits_per_sample": 5.9746,
+      "eval_cross_entropy_bits_per_sample": 5.5418,
+      "eval_nll_nats_per_sample": 3.8413
     },
     {
       "epoch": 8,
-      "train_loss_bits_per_sample": 5.530875523885091,
-      "eval_cross_entropy_bits_per_sample": 5.241023063659668
+      "train_loss_bits_per_sample": 5.8965,
+      "eval_cross_entropy_bits_per_sample": 5.2868,
+      "eval_nll_nats_per_sample": 3.6645
+    },
+    {
+      "epoch": 9,
+      "train_loss_bits_per_sample": 5.6948,
+      "eval_cross_entropy_bits_per_sample": 4.7803,
+      "eval_nll_nats_per_sample": 3.3135
+    },
+    {
+      "epoch": 10,
+      "train_loss_bits_per_sample": 5.4515,
+      "eval_cross_entropy_bits_per_sample": 4.0603,
+      "eval_nll_nats_per_sample": 2.8144
     }
   ],
   "exported_files": {
@@ -55,10 +78,10 @@
     "saved_model_dir": "pretrained_models/wavenet_waveform_coder/saved_model",
     "tflite": {
       "fp32_path": "pretrained_models/wavenet_waveform_coder/model_fp32.tflite",
-      "fp32_size_bytes": 2697196,
+      "fp32_size_bytes": 685348,
       "fp32_verified": true,
       "int8_path": "pretrained_models/wavenet_waveform_coder/model_int8.tflite",
-      "int8_size_bytes": 924192,
+      "int8_size_bytes": 181248,
       "int8_verified": true,
       "output_shape": [
         1,