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,
    "max_cached_steps": 2048,
    "dropout": 0.0,
    "conv_pos": 128,
    "conv_pos_groups": 16,
    "causal": false,
    "window_size": 512,
    "lookahead_size": 3,
    "use_flex_attention": false
  },
  "compressor_name": "FocalEncoder",
  "compressor_config": {
    "input_dim": 1024,
    "output_dim": 32,
    "hidden_dims": [
      1024,
      1024,
      1024
    ],
    "downscale_factors": [
      1,
      1,
      1
    ],
    "focal_window": 14,
    "focal_level": 2,
    "focal_factor": 4,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "tanhscale_init": 0.5,
    "normalize_modulator": false,
    "causal": false,
    "window_size": 512
  },
  "boundary_predictor_name": "HazardModel",
  "boundary_predictor_config": {
    "input_dim": 1024,
    "hidden_dims": [
      1024,
      1024,
      1024
    ],
    "downscale_factors": [
      1,
      1,
      1
    ],
    "focal_window": 14,
    "focal_level": 2,
    "focal_factor": 4,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "tanhscale_init": 0.5,
    "normalize_modulator": false,
    "causal": false,
    "window_size": 512
  },
  "downsampler_name": "SelectLastPool",
  "downsampler_config": {},
  "quantizer_name": "ScalarSphericalQuantizer",
  "quantizer_config": {
    "dim": 32,
    "n_levels": 4
  },
  "duration_predictor_name": "NegBinModel",
  "duration_predictor_config": {
    "input_dim": 32,
    "hidden_dims": [
      1024,
      1024,
      1024
    ],
    "downscale_factors": [
      1,
      1,
      1
    ],
    "focal_window": 14,
    "focal_level": 2,
    "focal_factor": 4,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "tanhscale_init": 0.5,
    "normalize_modulator": false,
    "causal": false,
    "window_size": 512,
    "min_duration": 1,
    "eps": 0.0001
  },
  "upsampler_name": "RepeatInterleaveUnpool",
  "upsampler_config": {},
  "decompressor_name": "FocalDecoder",
  "decompressor_config": {
    "input_dim": 32,
    "output_dim": 1024,
    "hidden_dims": [
      1024,
      1024,
      1024
    ],
    "upscale_factors": [
      1,
      1,
      1
    ],
    "focal_window": 14,
    "focal_level": 2,
    "focal_factor": 4,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "tanhscale_init": 0.5,
    "normalize_modulator": false,
    "causal": false,
    "window_size": 512,
    "last_window_size": 512,
    "lookahead_size": 3
  },
  "decoder_name": "Vocos",
  "decoder_config": {
    "input_dim": 1024,
    "num_layers": 8,
    "dim": 512,
    "ffn_dim": 1536,
    "kernel_size": 7,
    "layerscale_init": 0.125,
    "n_fft": 1024,
    "hop_length": 320,
    "causal": false
  },
  "char_aligner_name": "MMS",
  "char_aligner_config": {
    "checkpoint": "facebook/mms-1b-all"
  },
  "retriever_name": "LatentIVF",
  "retriever_config": {
    "input_dim": 1024,
    "latent_dim": 32,
    "hidden_dims": [
      1024,
      1024,
      1024
    ],
    "downscale_factors": [
      1,
      1,
      1
    ],
    "focal_window": 14,
    "focal_level": 2,
    "focal_factor": 4,
    "dropout": 0.0,
    "use_post_norm": false,
    "use_layerscale": false,
    "layerscale_init": 0.0001,
    "tanhscale_init": 0.5,
    "normalize_modulator": false,
    "causal": false,
    "window_size": 512,
    "nlist": 4096,
    "nprobe": 16
  }
}