{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 21.59,
    "wer_norm": 21.5,
    "wer_numcanon": 21.5,
    "space_norm_wer": 14.4,
    "mer": 9.81,
    "cer_norm": 9.76,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.09,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.1,
      "norm_to_mer": 11.69
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 20.8,
    "wer_norm": 20.35,
    "wer_numcanon": 20.35,
    "space_norm_wer": 12.91,
    "mer": 5.1,
    "cer_norm": 5.24,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.45,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.44,
      "norm_to_mer": 15.25
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 27.34,
    "wer_norm": 17.33,
    "wer_numcanon": 16.9,
    "space_norm_wer": 12.73,
    "mer": 12.53,
    "cer_norm": 12.02,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 10.01,
      "norm_to_numcanon": 0.43,
      "norm_to_space_norm": 4.6,
      "norm_to_mer": 4.8
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 19.84,
    "wer_norm": 19.82,
    "wer_numcanon": 19.82,
    "space_norm_wer": 14.8,
    "mer": 7.1,
    "cer_norm": 6.76,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 5.02,
      "norm_to_mer": 12.72
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 13.31,
    "wer_norm": 13.31,
    "wer_numcanon": 13.31,
    "space_norm_wer": 9.92,
    "mer": 4.9,
    "cer_norm": 4.61,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.39,
      "norm_to_mer": 8.41
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 43.16,
    "wer_norm": 43.16,
    "wer_numcanon": 43.16,
    "space_norm_wer": 35.26,
    "mer": 25.11,
    "cer_norm": 23.6,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.9,
      "norm_to_mer": 18.05
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 42.17,
    "wer_norm": 42.17,
    "wer_numcanon": 42.17,
    "space_norm_wer": 31.89,
    "mer": 9.08,
    "cer_norm": 9.06,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.28,
      "norm_to_mer": 33.09
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 25.18,
    "wer_norm": 25.18,
    "wer_numcanon": 25.18,
    "space_norm_wer": 18.4,
    "mer": 8.98,
    "cer_norm": 8.79,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.78,
      "norm_to_mer": 16.2
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 28.49,
    "wer_norm": 28.36,
    "wer_numcanon": 28.36,
    "space_norm_wer": 19.14,
    "mer": 10.13,
    "cer_norm": 9.9,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.13,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.22,
      "norm_to_mer": 18.23
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 23.18,
    "wer_norm": 22.83,
    "wer_numcanon": 22.83,
    "space_norm_wer": 19.59,
    "mer": 12.39,
    "cer_norm": 10.97,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.35,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.24,
      "norm_to_mer": 10.44
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 35.41,
    "wer_norm": 35.39,
    "wer_numcanon": 35.39,
    "space_norm_wer": 26.54,
    "mer": 12.53,
    "cer_norm": 12.2,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.85,
      "norm_to_mer": 22.86
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 27.12,
    "wer_norm": 27.1,
    "wer_numcanon": 27.1,
    "space_norm_wer": 16.56,
    "mer": 5.78,
    "cer_norm": 6.1,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.54,
      "norm_to_mer": 21.32
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 27.01,
    "wer_norm": 26.14,
    "wer_numcanon": 26.1,
    "space_norm_wer": 19.15,
    "mer": 10.61,
    "cer_norm": 10.21
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 27.3,
    "wer_norm": 26.38,
    "wer_numcanon": 26.34,
    "space_norm_wer": 19.34,
    "mer": 10.29,
    "cer_norm": 9.92
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/gemma3n-e2b-ckpt-40000",
    "checkpoint_name": "ckpt-40000",
    "model_id": "gemma3n-e2b",
    "model_type": "gemma3n-E2B-asr",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 128,
    "inference_time_sec": 598.31,
    "total_audio_sec": 40354.46,
    "rtf": 0.0148,
    "timestamp": "2026-03-29T07:53:51Z",
    "gpu": "NVIDIA A100-SXM4-80GB",
    "framework": "transformers",
    "normalization_version": "v1",
    "jiwer_version": "unknown"
  }
}