{
  "config": {"acoustic_width": 512, "width": 256, "vocabulary": 2048, "layers": 2, "heads": 4, "dropout": 0.1, "loops": 3},
  "memory": {"subkeys": 700, "key_dim": 64, "value_dim": 192, "context_frames": 3, "heads": 8, "top_k": 32, "temperature": 0.1, "dropout": 0.1},
  "learned_temperature": true,
  "subsampling_factor": 4,
  "language_head": true,
  "inference_loops": 3
}