egs/visinger2/config.json

{
  "train": {
    "log_interval": 200,
    "eval_interval": 1000,
    "seed": 1234,
    "port": 8001,
    "epochs": 10000,
    "learning_rate": 2e-4,
    "betas": [0.8, 0.99],
    "eps": 1e-9,
    "batch_size": 8,
    "accumulation_steps": 1,
    "fp16_run": false,
    "lr_decay": 0.998,
    "segment_size": 10240,
    "init_lr_ratio": 1,
    "warmup_epochs": 0,
    "c_mel": 45,
    "save_dir": "logdir/visinger2"
  },
  "data": {
    "data_dir":"../../data",
    "dataset_type": "SingDataset",
    "collate_type": "SingCollate",
    "training_filelist":"train.list",
    "training_labellist":"transcriptions.txt",
    "validation_filelist":"test.list",
    "validation_labellist":"transcriptions.txt",
    "max_wav_value": 32768.0,
    "sample_rate": 44100,
    "n_fft": 2048,
    "fmin": 0,
    "fmax": 22050,
    "hop_size": 512,
    "win_size": 2048,
    "acoustic_dim": 80,
    "min_level_db": -115,
    "ref_level_db": 20,
    "min_db": -115,
    "max_abs_value": 4.0,
    "n_speakers": 200,
    "spk2id": {"opencpop": 0, "taffy": 1, "otto": 2, "nanami": 3}
  },
  "model": {
    "hidden_channels": 192,
    "spk_channels": 192,
    "filter_channels": 768,
    "n_heads": 2,
    "n_layers": 4,
    "kernel_size": 3,
    "p_dropout": 0.1,
    "prior_hidden_channels": 192,
    "prior_filter_channels": 768,
    "prior_n_heads": 2,
    "prior_n_layers": 4,
    "prior_kernel_size": 3,
    "prior_p_dropout": 0.1,
    "resblock": "1",
    "use_spectral_norm": false,
    "resblock_kernel_sizes": [3,7,11],
    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
    "upsample_rates": [8,8,4,2],
    "upsample_initial_channel": 256,
    "upsample_kernel_sizes": [16,16,8,4],
    "n_harmonic": 64,
    "n_bands": 65
  }
}