config.json

{
  "engine": "bicodec",
  "components": {
    "wav2vec2_encoder": "wav2vec2_encoder.onnx",
    "wav2vec2_encoder_q8": "wav2vec2_encoder_q8.onnx",
    "semantic_encoder": "semantic_encoder.onnx",
    "semantic_encoder_q8": "semantic_encoder_q8.onnx",
    "global_encoder": "global_encoder.onnx",
    "global_encoder_q8": "global_encoder_q8.onnx",
    "mel_filterbank": "mel_filterbank.npy",
    "mel_config": "mel_config.json",
    "decoder": "decoder.onnx",
    "decoder_q8": "decoder_q8.onnx"
  },
  "sample_rates": {
    "input": 16000,
    "output": 16000
  },
  "distributable": true,
  "metadata": {
    "opset": 14,
    "sample_rate": 16000,
    "global_token_num": 32,
    "wav2vec2_layers": [
      11,
      14,
      16
    ],
    "vc_recipe": "source semantic tokens (content) + reference global tokens (timbre) \u2192 decode",
    "parity": {
      "wav2vec2_encoder": {
        "max_abs": 0.000671,
        "pass": true
      },
      "semantic_encoder": {
        "exact_int_match": true,
        "pass": true
      },
      "global_encoder": {
        "exact_int_match": true,
        "pass": true
      },
      "mel_numpy_vs_torchaudio": {
        "max_abs": 0.0,
        "pass": true
      },
      "decoder": {
        "max_abs": 2.5266781449317932e-06,
        "pass": true
      }
    }
  }
}