{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 21.35,
    "wer_norm": 21.14,
    "wer_numcanon": 21.14,
    "space_norm_wer": 13.65,
    "mer": 9.79,
    "cer_norm": 9.82,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.21,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.49,
      "norm_to_mer": 11.35
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 19.02,
    "wer_norm": 18.53,
    "wer_numcanon": 18.53,
    "space_norm_wer": 12.42,
    "mer": 4.94,
    "cer_norm": 4.97,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.49,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.11,
      "norm_to_mer": 13.59
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 28.16,
    "wer_norm": 17.61,
    "wer_numcanon": 17.61,
    "space_norm_wer": 12.84,
    "mer": 13.48,
    "cer_norm": 12.85,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 10.55,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.77,
      "norm_to_mer": 4.13
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 19.61,
    "wer_norm": 19.6,
    "wer_numcanon": 19.6,
    "space_norm_wer": 14.75,
    "mer": 7.68,
    "cer_norm": 7.23,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.01,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.85,
      "norm_to_mer": 11.92
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 13.5,
    "wer_norm": 13.5,
    "wer_numcanon": 13.5,
    "space_norm_wer": 10.16,
    "mer": 5.29,
    "cer_norm": 4.92,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.34,
      "norm_to_mer": 8.21
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 38.57,
    "wer_norm": 38.55,
    "wer_numcanon": 38.55,
    "space_norm_wer": 29.67,
    "mer": 18.95,
    "cer_norm": 18.14,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.88,
      "norm_to_mer": 19.6
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 39.98,
    "wer_norm": 39.96,
    "wer_numcanon": 39.96,
    "space_norm_wer": 30.27,
    "mer": 8.28,
    "cer_norm": 8.27,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.69,
      "norm_to_mer": 31.68
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 24.04,
    "wer_norm": 24.02,
    "wer_numcanon": 24.02,
    "space_norm_wer": 17.45,
    "mer": 7.75,
    "cer_norm": 7.7,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.57,
      "norm_to_mer": 16.27
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 30.23,
    "wer_norm": 30.1,
    "wer_numcanon": 30.1,
    "space_norm_wer": 20.5,
    "mer": 11.56,
    "cer_norm": 11.18,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.13,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.6,
      "norm_to_mer": 18.54
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 25.29,
    "wer_norm": 24.86,
    "wer_numcanon": 24.86,
    "space_norm_wer": 21.58,
    "mer": 15.11,
    "cer_norm": 13.25,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.43,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.28,
      "norm_to_mer": 9.75
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 34.35,
    "wer_norm": 34.35,
    "wer_numcanon": 34.35,
    "space_norm_wer": 25.24,
    "mer": 11.79,
    "cer_norm": 11.47,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.11,
      "norm_to_mer": 22.56
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 28.14,
    "wer_norm": 28.12,
    "wer_numcanon": 28.12,
    "space_norm_wer": 17.95,
    "mer": 7.58,
    "cer_norm": 7.65,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.17,
      "norm_to_mer": 20.54
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 26.58,
    "wer_norm": 25.64,
    "wer_numcanon": 25.64,
    "space_norm_wer": 18.68,
    "mer": 10.26,
    "cer_norm": 9.9
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 26.85,
    "wer_norm": 25.86,
    "wer_numcanon": 25.86,
    "space_norm_wer": 18.87,
    "mer": 10.18,
    "cer_norm": 9.79
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/gemma3n-e2b-ckpt-55000",
    "checkpoint_name": "ckpt-55000",
    "model_id": "gemma3n-e2b",
    "model_type": "gemma3n-E2B-asr",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 128,
    "inference_time_sec": 593.48,
    "total_audio_sec": 40354.46,
    "rtf": 0.0147,
    "timestamp": "2026-03-30T03:17:42Z",
    "gpu": "NVIDIA A100-SXM4-80GB",
    "framework": "transformers",
    "normalization_version": "v1",
    "jiwer_version": "unknown"
  }
}