{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 21.5,
    "wer_norm": 21.2,
    "wer_numcanon": 21.2,
    "space_norm_wer": 16.48,
    "mer": 9.06,
    "cer_norm": 8.79,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.3,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.72,
      "norm_to_mer": 12.14
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 20.45,
    "wer_norm": 19.98,
    "wer_numcanon": 19.98,
    "space_norm_wer": 13.2,
    "mer": 4.63,
    "cer_norm": 4.74,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.47,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.78,
      "norm_to_mer": 15.35
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 30.65,
    "wer_norm": 18.95,
    "wer_numcanon": 18.97,
    "space_norm_wer": 11.09,
    "mer": 17.18,
    "cer_norm": 16.34,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 11.7,
      "norm_to_numcanon": -0.02,
      "norm_to_space_norm": 7.86,
      "norm_to_mer": 1.77
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 20.97,
    "wer_norm": 20.97,
    "wer_numcanon": 20.97,
    "space_norm_wer": 15.93,
    "mer": 7.64,
    "cer_norm": 7.25,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 5.04,
      "norm_to_mer": 13.33
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 14.24,
    "wer_norm": 14.24,
    "wer_numcanon": 14.24,
    "space_norm_wer": 9.69,
    "mer": 4.61,
    "cer_norm": 4.49,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.55,
      "norm_to_mer": 9.63
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 73.51,
    "wer_norm": 73.43,
    "wer_numcanon": 71.0,
    "space_norm_wer": 55.06,
    "mer": 50.32,
    "cer_norm": 47.96,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.08,
      "norm_to_numcanon": 2.43,
      "norm_to_space_norm": 18.37,
      "norm_to_mer": 23.11
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 42.17,
    "wer_norm": 42.15,
    "wer_numcanon": 42.15,
    "space_norm_wer": 31.37,
    "mer": 8.14,
    "cer_norm": 8.25,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.78,
      "norm_to_mer": 34.01
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 25.75,
    "wer_norm": 25.75,
    "wer_numcanon": 25.75,
    "space_norm_wer": 18.22,
    "mer": 7.31,
    "cer_norm": 7.4,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.53,
      "norm_to_mer": 18.44
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 26.65,
    "wer_norm": 26.54,
    "wer_numcanon": 26.54,
    "space_norm_wer": 16.09,
    "mer": 6.5,
    "cer_norm": 6.88,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.11,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.45,
      "norm_to_mer": 20.04
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 27.28,
    "wer_norm": 26.9,
    "wer_numcanon": 26.9,
    "space_norm_wer": 21.62,
    "mer": 15.17,
    "cer_norm": 13.64,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.38,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 5.28,
      "norm_to_mer": 11.73
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 39.14,
    "wer_norm": 39.12,
    "wer_numcanon": 39.12,
    "space_norm_wer": 27.56,
    "mer": 15.04,
    "cer_norm": 14.74,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 11.56,
      "norm_to_mer": 24.08
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 27.59,
    "wer_norm": 27.59,
    "wer_numcanon": 27.59,
    "space_norm_wer": 17.11,
    "mer": 5.38,
    "cer_norm": 5.78,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.48,
      "norm_to_mer": 22.21
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 30.43,
    "wer_norm": 29.39,
    "wer_numcanon": 29.2,
    "space_norm_wer": 20.94,
    "mer": 13.21,
    "cer_norm": 12.73
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 30.82,
    "wer_norm": 29.73,
    "wer_numcanon": 29.53,
    "space_norm_wer": 21.12,
    "mer": 12.58,
    "cer_norm": 12.19
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/gemma3n-e2b-ckpt-20000",
    "checkpoint_name": "ckpt-20000",
    "model_id": "gemma3n-e2b",
    "model_type": "gemma3n-E2B-asr",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 128,
    "inference_time_sec": 702.81,
    "total_audio_sec": 40354.46,
    "rtf": 0.0174,
    "timestamp": "2026-03-29T08:06:16Z",
    "gpu": "NVIDIA A100-SXM4-80GB",
    "framework": "transformers",
    "normalization_version": "v1",
    "jiwer_version": "unknown"
  }
}