config.json

{
  "encoder_name": "WavLM",
  "encoder_config": {
    "hidden_dims": [512, 512, 512, 512, 512, 512, 512],
    "kernel_sizes": [10, 3, 3, 3, 3, 2, 2],
    "strides": [5, 2, 2, 2, 2, 2, 2],
    "num_layers": 6,
    "dim": 1024,
    "ffn_dim": 4096,
    "num_heads": 16,
    "num_buckets": 320,
    "max_distance": 800,
    "dropout": 0.0,
    "conv_pos": 128,
    "conv_pos_groups": 16
  },
  "compressor_name": "FocalEncoder",
  "compressor_config": {
    "input_dim": 1024,
    "output_dim": 13,
    "hidden_dims": [1024, 512, 256],
    "downscale_factors": [2, 1, 1],
    "focal_window": 7,
    "focal_level": 2,
    "focal_factor": 2,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "normalize_modulator": false
  },
  "quantizer_name": "BinarySphericalQuantizer",
  "quantizer_config": {
    "codebook_size": 8192
  },
  "decompressor_name": "FocalDecoder",
  "decompressor_config": {
    "input_dim": 13,
    "output_dim": 1024,
    "hidden_dims": [256, 512, 1024],
    "upscale_factors": [1, 1, 2],
    "focal_window": 7,
    "focal_level": 2,
    "focal_factor": 2,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "normalize_modulator": false
  },
  "decoder_name": "Vocos",
  "decoder_config": {
    "input_channels": 1024,
    "num_layers": 8,
    "dim": 512,
    "ffn_dim": 1536,
    "kernel_size": 7,
    "padding": 3,
    "layerscale_init": null,
    "n_fft": 1024,
    "hop_length": 320
  }
}