{
  "train_manifest": "resources/asr_memory_120s_mix/asr_memory_train.json",
  "base_enc_path": "siyuanTest/eval_speaker_embedding/ckpt/ywy_with_eos_0720_speech_encoder.pt",
  "source": "commonvoice",
  "speech_encoder_type": "transformer_v2",
  "audio_sample_rate": 16000,
  "layer_mode": "learned_mix",
  "layers": [
    5,
    7,
    8
  ],
  "input_dim": 1280,
  "projection_dim": 1024,
  "embedding_dim": 4096,
  "attention_dim": 256,
  "dropout": 0.15,
  "norm_type": "layernorm",
  "encoder_layer_truncation": true,
  "min_duration_sec": 1.5,
  "max_duration_sec": 10.0,
  "min_segments_per_speaker": 8,
  "speakers_per_batch": 32,
  "utterances_per_speaker": 3,
  "learning_rate": 0.0005,
  "min_learning_rate": 1e-05,
  "warmup_steps": 500,
  "aam_margin": 0.2,
  "aam_scale": 30.0,
  "supcon_weight": 0.1,
  "supcon_temperature": 0.07,
  "distributed_supcon_global": true,
  "num_train_speakers": 23328
}