{
  "dit": {
    "x_res": 384,
    "y_res": 256,
    "patch": 16,
    "dim": 384,
    "depth": 12,
    "heads": 6,
    "text_seq_dim": 768,
    "text_pool_dim": 512
  },
  "mel": {
    "sample_rate": 22050,
    "n_fft": 2048,
    "hop_length": 1024,
    "top_db": 80
  },
  "steps": 90000,
  "batch_size": 256,
  "train_pairs": 19195,
  "train_clips": 3839,
  "best_val": 0.06960614025592804
}