{
    "model_type": "Vocoder",
    "dataset": [
        "emilia"
    ],
    "preprocess": {
        "hop_size": 480,
        "sample_rate": 24000,
        "max_length": 36000,
        "n_fft": 1920,
        "num_mels": 128,
        "win_size": 1920,
        "fmin": 0,
        "fmax": 12000,
        "mel_var": 8.14,
        "mel_mean": -4.92,
        "processed_dir": "",
        "valid_file": "valid.json",
        "train_file": "train.json",
        "use_phone_cond": false,
        "use_emilia_101k": true
    },
    "model": {
        "vocos": {
            "input_channels": 128,
            "dim": 1024,
            "intermediate_dim": 4096,
            "num_layers": 30,
            "n_fft": 1920,
            "hop_size": 480,
            "padding": "same"
        },
        "period_gan": {
            "max_downsample_channels": 1024,
            "channels": 64,
            "channel_increasing_factor": 2
        },
        "spec_gan": {
            "stft_params": {
                "fft_sizes": [
                    128,
                    256,
                    512,
                    1024,
                    2048
                ],
                "hop_sizes": [
                    32,
                    64,
                    128,
                    256,
                    512
                ],
                "win_lengths": [
                    128,
                    256,
                    512,
                    1024,
                    2048
                ],
                "window": "hann_window"
            },
            "in_channels": 1,
            "out_channels": 1,
            "channels": 64,
            "kernel_sizes": [
                5,
                3
            ],
            "max_downsample_channels": 1024,
            "down_scales": [
                2,
                2,
                2
            ],
            "use_weight_norm": true,
            "use_complex": false
        }
    },
}