reference_transcripts_metadata.json

{
  "benchmark": {
    "name": "Voices for Christ Benchmark",
    "submission_process": "Submit a pull request that adds a new benchmark_data/<mode> directory with one output file per original filename.",
    "task": "speech restoration and denoising",
    "version": "0.3.0"
  },
  "dataset": {
    "clip_count": 100,
    "clip_duration_seconds": 30,
    "distribution_status": "public",
    "license_note": "Current working assumption: the source Voices for Christ recordings are pre-1990 and open-domain/public-domain compatible. This should be verified before wider redistribution or downstream commercial reuse.",
    "license_status": "assumed_public_domain_unverified",
    "name": "Voices for Christ archive excerpts",
    "source_format": "mp3",
    "splits": [
      "public_test"
    ],
    "storage": "git-lfs",
    "total_duration_seconds": 3000,
    "version": "0.1.0"
  },
  "reference_transcripts": {
    "asr_model": {
      "beam_size": 15,
      "best_of": 15,
      "compute_type": "float16",
      "condition_on_previous_text": true,
      "device": "cuda",
      "device_index": 0,
      "download_root": null,
      "language": "en",
      "model_name": "large-v3",
      "patience": 2.0
    },
    "policy": "Strong-reference decode of benchmark_data/original using a larger ASR model and wider beam than submission scoring."
  },
  "run": {
    "argv": [
      "score_asr_wer.py"
    ],
    "file_count": 100,
    "package_versions": {
      "faster-whisper": "1.2.0",
      "numpy": "2.2.6",
      "torch": "2.8.0",
      "torchaudio": "2.8.0",
      "tqdm": "4.67.3"
    },
    "platform": "Linux-6.5.0-15-generic-x86_64-with-glibc2.35",
    "python_version": "3.10.12",
    "timestamp_utc": "2026-03-17T22:58:48.810633Z"
  }
}