{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 14.25,
    "wer_norm": 14.06,
    "wer_numcanon": 14.06,
    "space_norm_wer": 11.83,
    "mer": 6.37,
    "cer_norm": 6.05,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.19,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 2.23,
      "norm_to_mer": 7.69
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 19.06,
    "wer_norm": 18.53,
    "wer_numcanon": 18.53,
    "space_norm_wer": 12.22,
    "mer": 4.75,
    "cer_norm": 4.78,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.53,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.31,
      "norm_to_mer": 13.78
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 33.57,
    "wer_norm": 23.92,
    "wer_numcanon": 23.95,
    "space_norm_wer": 19.4,
    "mer": 22.57,
    "cer_norm": 20.69,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 9.65,
      "norm_to_numcanon": -0.03,
      "norm_to_space_norm": 4.52,
      "norm_to_mer": 1.35
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 17.68,
    "wer_norm": 17.68,
    "wer_numcanon": 17.68,
    "space_norm_wer": 13.14,
    "mer": 6.16,
    "cer_norm": 5.89,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.54,
      "norm_to_mer": 11.52
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 12.31,
    "wer_norm": 12.31,
    "wer_numcanon": 12.31,
    "space_norm_wer": 9.09,
    "mer": 4.59,
    "cer_norm": 4.29,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.22,
      "norm_to_mer": 7.72
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 38.84,
    "wer_norm": 38.84,
    "wer_numcanon": 38.84,
    "space_norm_wer": 31.9,
    "mer": 21.11,
    "cer_norm": 19.88,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.94,
      "norm_to_mer": 17.73
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 39.13,
    "wer_norm": 39.13,
    "wer_numcanon": 39.13,
    "space_norm_wer": 29.64,
    "mer": 7.64,
    "cer_norm": 7.68,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.49,
      "norm_to_mer": 31.49
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 22.41,
    "wer_norm": 22.41,
    "wer_numcanon": 22.41,
    "space_norm_wer": 15.71,
    "mer": 6.3,
    "cer_norm": 6.39,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.7,
      "norm_to_mer": 16.11
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 33.55,
    "wer_norm": 33.35,
    "wer_numcanon": 33.35,
    "space_norm_wer": 24.25,
    "mer": 16.43,
    "cer_norm": 15.48,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.2,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.1,
      "norm_to_mer": 16.92
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 24.39,
    "wer_norm": 23.9,
    "wer_numcanon": 23.9,
    "space_norm_wer": 21.2,
    "mer": 15.33,
    "cer_norm": 13.29,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.49,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 2.7,
      "norm_to_mer": 8.57
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 32.14,
    "wer_norm": 32.12,
    "wer_numcanon": 32.12,
    "space_norm_wer": 23.55,
    "mer": 10.57,
    "cer_norm": 10.31,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.57,
      "norm_to_mer": 21.55
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 29.38,
    "wer_norm": 29.36,
    "wer_numcanon": 29.36,
    "space_norm_wer": 19.93,
    "mer": 9.99,
    "cer_norm": 9.82,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.43,
      "norm_to_mer": 19.37
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 25.91,
    "wer_norm": 25.03,
    "wer_numcanon": 25.03,
    "space_norm_wer": 19.03,
    "mer": 10.87,
    "cer_norm": 10.32
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 26.39,
    "wer_norm": 25.47,
    "wer_numcanon": 25.47,
    "space_norm_wer": 19.32,
    "mer": 10.98,
    "cer_norm": 10.38
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/gemma3n-e2b-ckpt-70000",
    "checkpoint_name": "ckpt-70000",
    "model_id": "gemma3n-e2b",
    "model_type": "gemma3n-E2B-asr",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 128,
    "inference_time_sec": 824.56,
    "total_audio_sec": 40354.46,
    "rtf": 0.0204,
    "timestamp": "2026-03-30T17:09:10Z",
    "gpu": "NVIDIA A100-SXM4-80GB",
    "framework": "transformers",
    "normalization_version": "v1",
    "jiwer_version": "unknown"
  }
}