{
  "audio_fr": 16000,
  "causal_attn": true,
  "class_dropout_prob": 0.1,
  "depth": 12,
  "grad_ckpt": false,
  "hidden_size": 768,
  "in_channels": 4,
  "latent_size": 32,
  "mlp_ratio": 4.0,
  "n_mels": 256,
  "num_classes": 0,
  "num_heads": 12,
  "patch_size": 1,
  "predict_frames": 10
}