{
  "architectures": [
    "SpeechDecoderModel"
  ],
  "backbone": "gru",
  "backbone_kwargs": {
    "bidirectional": false,
    "input_dropout": 0.2,
    "mlp_head": false,
    "n_layers": 5,
    "n_units": 768,
    "patch_size": 14,
    "patch_stride": 4,
    "rnn_dropout": 0.4
  },
  "dtype": "float32",
  "model_type": "neural-speech-decoder",
  "n_classes": 41,
  "n_days": 24,
  "n_phoneme_classes": 41,
  "neural_dim": 256,
  "prediction": "direct",
  "transformers_version": "5.5.3"
}