config.json

{
  "model_type": "gigaam-ctc",
  "base_model": "GigaAM v2 multilingual_ssl (220M, Conformer encoder)",
  "task": "automatic-speech-recognition",
  "language": "he",
  "sample_rate": 16000,
  "max_segment_seconds": 20,
  "head": "character CTC",
  "vocab_size": 68,
  "blank_id": 67,
  "training": {
    "stage_1": {"corpus": "SLPRL-HUJI/HebDB", "hours": 937, "labels": "weak", "epochs": 4, "lr": 1e-4},
    "stage_2": {"corpus": "ivrit-ai/crowd-transcribe-v5", "hours": 257, "labels": "human", "epochs": 4, "lr": 3e-5}
  },
  "results": {
    "crowd_transcribe_v5_1000_clips_wer": 13.60,
    "eval_whatsapp_wer": 14.71
  }
}