{
  "model": "VoiceHub/DACVAE-Turkish",
  "training_source_commit": "ef31744",
  "github_report_commit": "d82893caf3b86ba58b8947caddf1a8ce205397f5",
  "checkpoint_sha256": "9f0db3e4eb7b5f242222926df4ac64b72ed8e1c67c07e2f5f87e8f6b14ff4f96",
  "weights_sha256": "9f2ca66004a7457f10e105c0eba3f8ce44961e2f9ef65bb2813ddbd2e464c7dd",
  "evaluation": {
    "split": "validation",
    "crop_samples": 256,
    "crop_seconds": 3,
    "asr_full_utterances": 32,
    "asr_duration_seconds": 224.24159963151928,
    "sample_rate": 48000,
    "codec_precision": "fp32",
    "posterior": "mean",
    "watermark": "bypass",
    "loudness_db": -16,
    "asr_model": "openai/whisper-large-v3-turbo",
    "asr_revision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9"
  },
  "models": {
    "meta": {
      "reconstruction": {
        "mel": 0.2786256688996218,
        "stft": 1.0640862321015447,
        "si_sdr_db": 16.884010054171085
      },
      "asr": {
        "wer": {
          "errors": 23,
          "substitutions": 17,
          "deletions": 4,
          "insertions": 2,
          "reference_units": 461,
          "hypothesis_units": 459,
          "rate": 0.049891540130151846
        },
        "cer": {
          "errors": 74,
          "substitutions": 24,
          "deletions": 32,
          "insertions": 18,
          "reference_units": 2886,
          "hypothesis_units": 2872,
          "rate": 0.02564102564102564
        }
      }
    },
    "japanese": {
      "reconstruction": {
        "mel": 0.26558971946360543,
        "stft": 0.971278426470235,
        "si_sdr_db": 17.442306242883205
      },
      "asr": {
        "wer": {
          "errors": 23,
          "substitutions": 18,
          "deletions": 4,
          "insertions": 1,
          "reference_units": 461,
          "hypothesis_units": 458,
          "rate": 0.049891540130151846
        },
        "cer": {
          "errors": 73,
          "substitutions": 22,
          "deletions": 34,
          "insertions": 17,
          "reference_units": 2886,
          "hypothesis_units": 2869,
          "rate": 0.025294525294525295
        }
      }
    },
    "turkish": {
      "reconstruction": {
        "mel": 0.22691508114803582,
        "stft": 0.8432678864337504,
        "si_sdr_db": 17.419028524309397
      },
      "asr": {
        "wer": {
          "errors": 19,
          "substitutions": 15,
          "deletions": 3,
          "insertions": 1,
          "reference_units": 461,
          "hypothesis_units": 459,
          "rate": 0.04121475054229935
        },
        "cer": {
          "errors": 66,
          "substitutions": 18,
          "deletions": 34,
          "insertions": 14,
          "reference_units": 2886,
          "hypothesis_units": 2866,
          "rate": 0.02286902286902287
        }
      }
    }
  },
  "paired_episode_bootstrap": {
    "meta": {
      "method": "paired episode-cluster percentile bootstrap",
      "estimand": "pooled sample mean; episodes resampled uniformly with all their paired rows",
      "n_samples": 256,
      "n_episodes": 18,
      "resamples": 10000,
      "confidence": 0.95,
      "metrics": {
        "mel": {
          "lower_is_better": true,
          "baseline_mean": 0.2786256688996218,
          "candidate_mean": 0.22691508114803582,
          "candidate_minus_baseline": -0.05171058775158599,
          "delta_ci": [
            -0.0592170431373271,
            -0.04379978859163707
          ],
          "sample_improvement_fraction": 1.0,
          "sample_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "relative_loss_reduction_percent": 18.559161456949372,
          "relative_loss_reduction_percent_ci": [
            15.841186723179938,
            21.090114311096205
          ]
        },
        "stft": {
          "lower_is_better": true,
          "baseline_mean": 1.0640862321015447,
          "candidate_mean": 0.8432678864337504,
          "candidate_minus_baseline": -0.22081834566779435,
          "delta_ci": [
            -0.2577764498334056,
            -0.18161389496220354
          ],
          "sample_improvement_fraction": 1.0,
          "sample_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "relative_loss_reduction_percent": 20.751922072300797,
          "relative_loss_reduction_percent_ci": [
            17.674230372339018,
            23.3378993907297
          ]
        },
        "si_sdr_db": {
          "lower_is_better": false,
          "baseline_mean": 16.884010054171085,
          "candidate_mean": 17.419028524309397,
          "candidate_minus_baseline": 0.5350184701383114,
          "delta_ci": [
            0.4437410553231082,
            0.6316349267443525
          ],
          "sample_improvement_fraction": 0.91796875,
          "sample_improvement_fraction_ci": [
            0.854251012145749,
            0.9700854700854701
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ]
        },
        "mel_plus_stft": {
          "lower_is_better": true,
          "baseline_mean": 1.3427119010011666,
          "candidate_mean": 1.0701829675817862,
          "candidate_minus_baseline": -0.27252893341938034,
          "delta_ci": [
            -0.3139729544149179,
            -0.22711216552550356
          ],
          "sample_improvement_fraction": 1.0,
          "sample_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "relative_loss_reduction_percent": 20.29690309709585,
          "relative_loss_reduction_percent_ci": [
            17.405983380223457,
            22.672448447772194
          ]
        }
      }
    },
    "japanese": {
      "method": "paired episode-cluster percentile bootstrap",
      "estimand": "pooled sample mean; episodes resampled uniformly with all their paired rows",
      "n_samples": 256,
      "n_episodes": 18,
      "resamples": 10000,
      "confidence": 0.95,
      "metrics": {
        "mel": {
          "lower_is_better": true,
          "baseline_mean": 0.26558971946360543,
          "candidate_mean": 0.22691508114803582,
          "candidate_minus_baseline": -0.03867463831556961,
          "delta_ci": [
            -0.043182247805592405,
            -0.03414554283863557
          ],
          "sample_improvement_fraction": 1.0,
          "sample_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "relative_loss_reduction_percent": 14.561797946727118,
          "relative_loss_reduction_percent_ci": [
            12.812924720810779,
            16.33984431013747
          ]
        },
        "stft": {
          "lower_is_better": true,
          "baseline_mean": 0.971278426470235,
          "candidate_mean": 0.8432678864337504,
          "candidate_minus_baseline": -0.1280105400364846,
          "delta_ci": [
            -0.14295202443047406,
            -0.1122004633242202
          ],
          "sample_improvement_fraction": 1.0,
          "sample_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "relative_loss_reduction_percent": 13.179592642831906,
          "relative_loss_reduction_percent_ci": [
            11.664869168390986,
            14.588833521267814
          ]
        },
        "si_sdr_db": {
          "lower_is_better": false,
          "baseline_mean": 17.442306242883205,
          "candidate_mean": 17.419028524309397,
          "candidate_minus_baseline": -0.02327771857380867,
          "delta_ci": [
            -0.0824172738416143,
            0.036898595112216714
          ],
          "sample_improvement_fraction": 0.3828125,
          "sample_improvement_fraction_ci": [
            0.2766798418972332,
            0.4947791811846689
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 0.3888888888888889,
          "episode_improvement_fraction_ci": [
            0.16666666666666666,
            0.6111111111111112
          ]
        },
        "mel_plus_stft": {
          "lower_is_better": true,
          "baseline_mean": 1.2368681459338404,
          "candidate_mean": 1.0701829675817862,
          "candidate_minus_baseline": -0.1666851783520542,
          "delta_ci": [
            -0.18402446937180994,
            -0.14767500246274123
          ],
          "sample_improvement_fraction": 1.0,
          "sample_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "sample_tie_fraction": 0.0,
          "episode_improvement_fraction": 1.0,
          "episode_improvement_fraction_ci": [
            1.0,
            1.0
          ],
          "relative_loss_reduction_percent": 13.476390260354407,
          "relative_loss_reduction_percent_ci": [
            11.990223045831089,
            14.795814532473681
          ]
        }
      }
    }
  },
  "limitations": [
    "Single training seed for waveform150",
    "Development validation used for selection; not final test",
    "Only two ASR utterances improve versus baselines",
    "No listening/MOS/speaker/TTS superiority claim",
    "Turkish model fine-tuned from Japanese, not directly from Meta"
  ]
}