{
  "Bengali": {
    "n_samples": 2236,
    "wer_raw": 14.19,
    "wer_norm": 14.17,
    "wer_numcanon": 14.17,
    "space_norm_wer": 9.44,
    "mer": 2.74,
    "cer_norm": 2.85,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.03,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.73,
      "norm_to_mer": 11.43
    }
  },
  "Gujarati": {
    "n_samples": 2905,
    "wer_raw": 14.92,
    "wer_norm": 14.92,
    "wer_numcanon": 14.92,
    "space_norm_wer": 10.49,
    "mer": 3.31,
    "cer_norm": 3.33,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.43,
      "norm_to_mer": 11.61
    }
  },
  "Hindi": {
    "n_samples": 3151,
    "wer_raw": 13.05,
    "wer_norm": 13.05,
    "wer_numcanon": 13.05,
    "space_norm_wer": 9.44,
    "mer": 3.81,
    "cer_norm": 3.66,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.6,
      "norm_to_mer": 9.24
    }
  },
  "Kannada": {
    "n_samples": 2062,
    "wer_raw": 20.93,
    "wer_norm": 20.93,
    "wer_numcanon": 20.93,
    "space_norm_wer": 13.14,
    "mer": 3.36,
    "cer_norm": 3.68,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.78,
      "norm_to_mer": 17.57
    }
  },
  "Malayalam": {
    "n_samples": 1767,
    "wer_raw": 39.36,
    "wer_norm": 39.36,
    "wer_numcanon": 39.36,
    "space_norm_wer": 30.16,
    "mer": 7.14,
    "cer_norm": 7.16,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.21,
      "norm_to_mer": 32.22
    }
  },
  "Marathi": {
    "n_samples": 2378,
    "wer_raw": 17.06,
    "wer_norm": 17.06,
    "wer_numcanon": 17.06,
    "space_norm_wer": 11.98,
    "mer": 3.8,
    "cer_norm": 3.91,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 5.07,
      "norm_to_mer": 13.25
    }
  },
  "Odia": {
    "n_samples": 2397,
    "wer_raw": 20.54,
    "wer_norm": 20.5,
    "wer_numcanon": 20.5,
    "space_norm_wer": 12.65,
    "mer": 3.68,
    "cer_norm": 3.94,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.04,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.85,
      "norm_to_mer": 16.82
    }
  },
  "Punjabi": {
    "n_samples": 3256,
    "wer_raw": 15.36,
    "wer_norm": 14.92,
    "wer_numcanon": 14.92,
    "space_norm_wer": 11.49,
    "mer": 4.31,
    "cer_norm": 4.05,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.44,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.44,
      "norm_to_mer": 10.61
    }
  },
  "Tamil": {
    "n_samples": 2770,
    "wer_raw": 26.69,
    "wer_norm": 26.63,
    "wer_numcanon": 26.63,
    "space_norm_wer": 18.32,
    "mer": 4.67,
    "cer_norm": 4.86,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.06,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.31,
      "norm_to_mer": 21.96
    }
  },
  "Telugu": {
    "n_samples": 2379,
    "wer_raw": 24.16,
    "wer_norm": 24.16,
    "wer_numcanon": 24.16,
    "space_norm_wer": 15.07,
    "mer": 4.09,
    "cer_norm": 4.45,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.09,
      "norm_to_mer": 20.07
    }
  },
  "__overall__": {
    "n_samples": 25301,
    "wer_raw": 19.63,
    "wer_norm": 19.56,
    "wer_numcanon": 19.56,
    "space_norm_wer": 13.54,
    "mer": 4.17,
    "cer_norm": 4.25
  },
  "__macro_avg__": {
    "n_languages": 10,
    "wer_raw": 20.63,
    "wer_norm": 20.57,
    "wer_numcanon": 20.57,
    "space_norm_wer": 14.22,
    "mer": 4.09,
    "cer_norm": 4.19
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/qwen3-asr-mixed-v2-ckpt-350000",
    "checkpoint_name": "ckpt-350000",
    "model_id": "qwen3-asr-mixed-v2-kathbath",
    "model_type": "qwen3-asr-1.7B-mixed-v2",
    "dataset": "ai4bharat/Kathbath",
    "batch_size": 128,
    "inference_time_sec": 2305.97,
    "total_audio_sec": 167150.92,
    "rtf": 0.0138,
    "timestamp": "2026-04-07T09:45:06Z",
    "framework": "transformers",
    "normalization_version": "v1"
  }
}