{
  "assamese": {
    "n_samples": 500,
    "wer_raw": 49.22,
    "wer_norm": 49.04,
    "wer_numcanon": 49.04,
    "space_norm_wer": 41.39,
    "mer": 23.87,
    "cer_norm": 22.36,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.18,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.65,
      "norm_to_mer": 25.17
    }
  },
  "bengali": {
    "n_samples": 500,
    "wer_raw": 31.43,
    "wer_norm": 30.79,
    "wer_numcanon": 30.79,
    "space_norm_wer": 22.9,
    "mer": 10.17,
    "cer_norm": 9.91,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.64,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.89,
      "norm_to_mer": 20.62
    }
  },
  "english": {
    "n_samples": 500,
    "wer_raw": 26.99,
    "wer_norm": 14.27,
    "wer_numcanon": 13.79,
    "space_norm_wer": 8.27,
    "mer": 8.69,
    "cer_norm": 8.72,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 12.72,
      "norm_to_numcanon": 0.48,
      "norm_to_space_norm": 6.0,
      "norm_to_mer": 5.58
    }
  },
  "gujarati": {
    "n_samples": 500,
    "wer_raw": 26.74,
    "wer_norm": 26.36,
    "wer_numcanon": 26.36,
    "space_norm_wer": 20.67,
    "mer": 8.9,
    "cer_norm": 8.68,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.38,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 5.69,
      "norm_to_mer": 17.46
    }
  },
  "hindi": {
    "n_samples": 500,
    "wer_raw": 14.31,
    "wer_norm": 14.14,
    "wer_numcanon": 14.12,
    "space_norm_wer": 10.99,
    "mer": 5.35,
    "cer_norm": 4.95,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.17,
      "norm_to_numcanon": 0.02,
      "norm_to_space_norm": 3.15,
      "norm_to_mer": 8.79
    }
  },
  "kannada": {
    "n_samples": 500,
    "wer_raw": 46.03,
    "wer_norm": 45.25,
    "wer_numcanon": 45.25,
    "space_norm_wer": 34.16,
    "mer": 11.8,
    "cer_norm": 12.09,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.78,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 11.09,
      "norm_to_mer": 33.45
    }
  },
  "malayalam": {
    "n_samples": 500,
    "wer_raw": 51.75,
    "wer_norm": 50.57,
    "wer_numcanon": 50.57,
    "space_norm_wer": 40.55,
    "mer": 12.36,
    "cer_norm": 12.15,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 1.18,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.02,
      "norm_to_mer": 38.21
    }
  },
  "marathi": {
    "n_samples": 500,
    "wer_raw": 38.99,
    "wer_norm": 38.67,
    "wer_numcanon": 38.67,
    "space_norm_wer": 30.37,
    "mer": 12.69,
    "cer_norm": 12.64,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.32,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 8.3,
      "norm_to_mer": 25.98
    }
  },
  "odia": {
    "n_samples": 500,
    "wer_raw": 40.71,
    "wer_norm": 40.06,
    "wer_numcanon": 40.06,
    "space_norm_wer": 30.21,
    "mer": 12.91,
    "cer_norm": 12.67,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.65,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 9.85,
      "norm_to_mer": 27.15
    }
  },
  "punjabi": {
    "n_samples": 500,
    "wer_raw": 28.82,
    "wer_norm": 28.46,
    "wer_numcanon": 28.46,
    "space_norm_wer": 20.56,
    "mer": 12.63,
    "cer_norm": 12.09,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 0.36,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 7.9,
      "norm_to_mer": 15.83
    }
  },
  "tamil": {
    "n_samples": 500,
    "wer_raw": 48.56,
    "wer_norm": 47.17,
    "wer_numcanon": 47.17,
    "space_norm_wer": 36.3,
    "mer": 12.23,
    "cer_norm": 12.25,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 1.39,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 10.87,
      "norm_to_mer": 34.94
    }
  },
  "telugu": {
    "n_samples": 500,
    "wer_raw": 41.07,
    "wer_norm": 39.54,
    "wer_numcanon": 39.54,
    "space_norm_wer": 27.42,
    "mer": 9.52,
    "cer_norm": 9.83,
    "empty_hypotheses": 0,
    "normalization_delta": {
      "raw_to_norm": 1.53,
      "norm_to_numcanon": 0.0,
      "norm_to_space_norm": 12.12,
      "norm_to_mer": 30.02
    }
  },
  "__overall__": {
    "n_samples": 6000,
    "wer_raw": 37.2,
    "wer_norm": 35.59,
    "wer_numcanon": 35.55,
    "space_norm_wer": 27.27,
    "mer": 12.05,
    "cer_norm": 11.84
  },
  "__macro_avg__": {
    "n_languages": 12,
    "wer_raw": 37.05,
    "wer_norm": 35.36,
    "wer_numcanon": 35.32,
    "space_norm_wer": 26.98,
    "mer": 11.76,
    "cer_norm": 11.53
  },
  "__meta__": {
    "checkpoint": "/home/ubuntu/training/checkpoints/qwen3-asr-ckpt-170000",
    "checkpoint_name": "ckpt-170000",
    "model_id": "qwen3-asr",
    "model_type": "Qwen3-ASR-1.7B",
    "dataset": "BayAreaBoys/indic-asr-benchmark-6k",
    "batch_size": 64,
    "inference_time_sec": 269.19,
    "total_audio_sec": 40354.46,
    "rtf": 0.0067,
    "timestamp": "2026-03-28T06:50:37Z",
    "gpu": "NVIDIA H200 80GB",
    "framework": "vllm",
    "normalization_version": "v1",
    "jiwer_version": "3.1.0"
  }
}