{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 29.06,
    "wer_norm": 28.91,
    "wer_numcanon": 28.91,
    "space_norm_wer": 24.53,
    "mer": 13.12,
    "cer_norm": 12.34,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.15,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.38,
      "norm_to_mer": 15.79
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 23.63,
    "wer_norm": 22.87,
    "wer_numcanon": 22.87,
    "space_norm_wer": 15.96,
    "mer": 6.44,
    "cer_norm": 6.5,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.76,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.91,
      "norm_to_mer": 16.43
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 27.25,
    "wer_norm": 15.37,
    "wer_numcanon": 14.91,
    "space_norm_wer": 9.22,
    "mer": 9.28,
    "cer_norm": 9.22,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 11.88,
      "norm_to_numcanon": 0.46,
      "norm_to_space_norm": 6.15,
      "norm_to_mer": 6.09
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 20.81,
    "wer_norm": 20.54,
    "wer_numcanon": 20.54,
    "space_norm_wer": 15.18,
    "mer": 6.27,
    "cer_norm": 6.16,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.27,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 5.36,
      "norm_to_mer": 14.27
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 13.93,
    "wer_norm": 13.88,
    "wer_numcanon": 13.88,
    "space_norm_wer": 10.75,
    "mer": 5.11,
    "cer_norm": 4.74,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.05,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 3.13,
      "norm_to_mer": 8.77
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 34.66,
    "wer_norm": 34.03,
    "wer_numcanon": 34.03,
    "space_norm_wer": 24.35,
    "mer": 7.92,
    "cer_norm": 8.25,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.63,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.68,
      "norm_to_mer": 26.11
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 44.23,
    "wer_norm": 43.49,
    "wer_numcanon": 43.49,
    "space_norm_wer": 33.21,
    "mer": 8.99,
    "cer_norm": 9.0,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.74,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.28,
      "norm_to_mer": 34.5
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 29.18,
    "wer_norm": 28.95,
    "wer_numcanon": 28.95,
    "space_norm_wer": 22.16,
    "mer": 9.05,
    "cer_norm": 9.07,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.23,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 6.79,
      "norm_to_mer": 19.9
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 27.61,
    "wer_norm": 27.35,
    "wer_numcanon": 27.35,
    "space_norm_wer": 18.05,
    "mer": 7.02,
    "cer_norm": 7.26,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.26,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.3,
      "norm_to_mer": 20.33
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 21.57,
    "wer_norm": 21.32,
    "wer_numcanon": 21.32,
    "space_norm_wer": 16.35,
    "mer": 7.05,
    "cer_norm": 6.75,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.25,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 4.97,
      "norm_to_mer": 14.27
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 40.3,
    "wer_norm": 39.12,
    "wer_numcanon": 39.12,
    "space_norm_wer": 28.82,
    "mer": 9.24,
    "cer_norm": 9.38,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 1.18,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.3,
      "norm_to_mer": 29.88
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 32.54,
    "wer_norm": 31.54,
    "wer_numcanon": 31.54,
    "space_norm_wer": 20.13,
    "mer": 6.51,
    "cer_norm": 6.89,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 1.0,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 11.41,
      "norm_to_mer": 25.03
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 28.59,
    "wer_norm": 27.21,
    "wer_numcanon": 27.17,
    "space_norm_wer": 19.95,
    "mer": 8.18,
    "cer_norm": 8.16
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 28.73,
    "wer_norm": 27.28,
    "wer_numcanon": 27.24,
    "space_norm_wer": 19.89,
    "mer": 8.0,
    "cer_norm": 7.96
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/qwen3-asr-ckpt-400000",
    "checkpoint_name": "ckpt-400000",
    "model_id": "qwen3-asr",
    "model_type": "Qwen3-ASR-1.7B",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 64,
    "inference_time_sec": 265.91,
    "total_audio_sec": 40354.46,
    "rtf": 0.0066,
    "timestamp": "2026-03-29T07:15:16Z",
    "gpu": "NVIDIA H200 80GB",
    "framework": "vllm",
    "normalization_version": "v1",
    "jiwer_version": "3.1.0"
  }
}