{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 15.15,
    "wer_norm": 14.97,
    "wer_numcanon": 14.97,
    "space_norm_wer": 12.47,
    "mer": 6.27,
    "cer_norm": 5.94,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.18,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 2.5,
      "norm_to_mer": 8.7
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 16.16,
    "wer_norm": 15.66,
    "wer_numcanon": 15.66,
    "space_norm_wer": 9.57,
    "mer": 2.69,
    "cer_norm": 2.92,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.49,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.09,
      "norm_to_mer": 12.98
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 27.45,
    "wer_norm": 12.47,
    "wer_numcanon": 12.47,
    "space_norm_wer": 7.88,
    "mer": 8.35,
    "cer_norm": 8.2,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 14.99,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.58,
      "norm_to_mer": 4.11
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 15.26,
    "wer_norm": 15.26,
    "wer_numcanon": 15.26,
    "space_norm_wer": 10.71,
    "mer": 3.6,
    "cer_norm": 3.64,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.56,
      "norm_to_mer": 11.66
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 13.03,
    "wer_norm": 13.03,
    "wer_numcanon": 13.03,
    "space_norm_wer": 9.05,
    "mer": 3.89,
    "cer_norm": 3.79,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.98,
      "norm_to_mer": 9.15
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 24.33,
    "wer_norm": 24.28,
    "wer_numcanon": 24.28,
    "space_norm_wer": 15.71,
    "mer": 4.2,
    "cer_norm": 4.62,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.05,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.57,
      "norm_to_mer": 20.08
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 38.86,
    "wer_norm": 38.86,
    "wer_numcanon": 38.86,
    "space_norm_wer": 29.39,
    "mer": 6.87,
    "cer_norm": 6.94,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.47,
      "norm_to_mer": 31.99
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 20.98,
    "wer_norm": 20.98,
    "wer_numcanon": 20.98,
    "space_norm_wer": 14.53,
    "mer": 5.18,
    "cer_norm": 5.38,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.45,
      "norm_to_mer": 15.8
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 22.68,
    "wer_norm": 22.61,
    "wer_numcanon": 22.61,
    "space_norm_wer": 13.01,
    "mer": 3.91,
    "cer_norm": 4.39,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.07,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.6,
      "norm_to_mer": 18.7
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 15.62,
    "wer_norm": 15.14,
    "wer_numcanon": 15.14,
    "space_norm_wer": 11.78,
    "mer": 4.29,
    "cer_norm": 4.05,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.49,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.36,
      "norm_to_mer": 10.85
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 27.93,
    "wer_norm": 27.91,
    "wer_numcanon": 27.91,
    "space_norm_wer": 18.74,
    "mer": 4.99,
    "cer_norm": 5.22,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.02,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.17,
      "norm_to_mer": 22.92
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 22.72,
    "wer_norm": 22.72,
    "wer_numcanon": 22.72,
    "space_norm_wer": 13.73,
    "mer": 3.76,
    "cer_norm": 4.17,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.99,
      "norm_to_mer": 18.96
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 21.38,
    "wer_norm": 20.09,
    "wer_numcanon": 20.09,
    "space_norm_wer": 13.8,
    "mer": 4.91,
    "cer_norm": 5.02
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 21.68,
    "wer_norm": 20.32,
    "wer_numcanon": 20.32,
    "space_norm_wer": 13.88,
    "mer": 4.83,
    "cer_norm": 4.94
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/qwen3-asr-mixed-v2-ckpt-450000",
    "checkpoint_name": "ckpt-450000",
    "model_id": "qwen3-asr-mixed-v2",
    "model_type": "qwen3-asr-1.7B-mixed-v2",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 128,
    "inference_time_sec": 677.77,
    "total_audio_sec": 40354.46,
    "rtf": 0.0168,
    "timestamp": "2026-04-08T21:20:22Z",
    "gpu": "NVIDIA A100-SXM4-80GB",
    "framework": "transformers",
    "normalization_version": "v1"
  }
}