config.json

{
  "_class_name": "MotifAudio",
  "auto_map": {
    "AutoModel": "model.MotifAudio"
  },
  "sample_rate": 16000,
  "semantic_config": {
    "norm_type": "global",
    "norm_mean": -7.115164,
    "norm_std": 4.229471,
    "sem_dim": 1024,
    "n_mels": 80,
    "n_fft": 400,
    "hop_length": 160,
    "f_min": 50.0,
    "f_max": 8000.0,
    "patch_f": 16,
    "patch_t": 16,
    "semantic_encoder": {
      "img_size": [
        80,
        512
      ],
      "patch_size": [
        16,
        16
      ],
      "in_chans": 1,
      "embed_dim": 1024,
      "depth": 48,
      "num_heads": 16,
      "mlp_ratio": 4.0,
      "use_cls_token": false,
      "num_register_tokens": 4,
      "use_rope": true,
      "rope_theta": 100.0,
      "layer_scale_init": 0.0
    }
  },
  "acoustic_config": {
    "sample_rate": 16000,
    "n_fft": 1024,
    "hop_length": 160,
    "n_mels": 160,
    "patch_f": 160,
    "patch_t": 2,
    "embed_dim": 1024,
    "f_min": 50.0,
    "f_max": 8000.0,
    "norm_mean": -7.115164,
    "norm_std": 4.229471
  },
  "fusion_config": {
    "dim": 1024,
    "num_heads": 8
  },
  "decoder_config": {
    "input_channels": 1024,
    "hidden_dim": 1024,
    "intermediate_dim": 4096,
    "num_layers": 12,
    "n_fft": 1024,
    "hop_length": 160,
    "sample_rate": 16000,
    "upsample_tokens": 2
  }
}