{
  "encoder_name": "LogMelSpectrogram",
  "encoder_config": {
    "sample_rate": 16000,
    "n_fft": 400,
    "win_length": 400,
    "hop_length": 160,
    "n_mels": 80,
    "f_min": 0.0,
    "f_max": 8000.0,
    "eps": 1e-06,
    "stft_backend": "conv"
  },
  "frontend_name": "Patch1d",
  "frontend_config": {
    "input_dim": 80,
    "output_dim": 2048,
    "patch_size": 16
  },
  "compressor_name": "ErfFormer",
  "compressor_config": {
    "dim": 2048,
    "ffn_dim": 8192,
    "num_layers": 6,
    "num_heads": 16,
    "num_kv_heads": 4,
    "dropout": 0.0,
    "window_size": 256,
    "erfscale_init": 0.5,
    "shift_init": 0.0,
    "native_gqa": false
  },
  "quantizer_name": "ScalarSphericalQuantizer",
  "quantizer_config": {
    "input_dim": 2048,
    "latent_dim": 64,
    "bits_per_dim": 2,
    "erfscale_init": 0.5,
    "shift_init": 0.0
  },
  "decompressor_name": "ErfFormer",
  "decompressor_config": {
    "dim": 2048,
    "ffn_dim": 8192,
    "num_layers": 6,
    "num_heads": 16,
    "num_kv_heads": 4,
    "dropout": 0.0,
    "window_size": 256,
    "erfscale_init": 0.5,
    "shift_init": 0.0,
    "native_gqa": false
  },
  "backend_name": "Unpatch1d",
  "backend_config": {
    "input_dim": 2048,
    "output_dim": 1024,
    "patch_size": 8
  },
  "decoder_name": "Vocos",
  "decoder_config": {
    "input_dim": 1024,
    "num_layers": 20,
    "dim": 1024,
    "ffn_dim": 2048,
    "kernel_size": 7,
    "patch_size": 8,
    "n_fft": 1024,
    "win_length": 1024,
    "hop_length": 320,
    "istft_backend": "conv",
    "eps": 1e-06
  }
}