config/vocoder.json

{
  "base_config": "config/base.json",
  "dataset": [
    "LJSpeech",
    "LibriTTS",
    "opencpop",
    "m4singer",
    "svcc",
    "svcceval",
    "pjs",
    "opensinger",
    "popbutfy",
    "nus48e",
    "popcs",
    "kising",
    "csd",
    "opera",
    "vctk",
    "lijian",
    "cdmusiceval"
  ],
  "task_type": "vocoder",
  "preprocess": {
    // acoustic features
    "extract_mel": true,
    "extract_pitch": false,
    "extract_uv": false,
    "extract_audio": true,
    "extract_label": false,
    "extract_one_hot": false,
    "extract_amplitude_phase": false,
    "pitch_extractor": "parselmouth",
    // Settings for data preprocessing
    "n_mel": 100,
    "win_size": 1024,
    "hop_size": 256,
    "sample_rate": 24000,
    "n_fft": 1024,
    "fmin": 0,
    "fmax": 12000,
    "f0_min": 50,
    "f0_max": 1100,
    "pitch_bin": 256,
    "pitch_max": 1100.0,
    "pitch_min": 50.0,
    "is_mu_law": false,
    "bits": 8,
    "cut_mel_frame": 32,
    // Directory names of processed data or extracted features
    "spk2id": "singers.json",
    // Features used for model training
    "use_mel": true,
    "use_frame_pitch": false,
    "use_uv": false,
    "use_audio": true,
    "use_label": false,
    "use_one_hot": false,
    "train_file": "train.json",
    "valid_file": "test.json"
  },
  "train": {
    "random_seed": 114514,
    "batch_size": 64,
    "gradient_accumulation_step": 1,
    "max_epoch": 1000000,
    "save_checkpoint_stride": [
      20
    ],
    "run_eval": [
      true
    ],
    "sampler": {
      "holistic_shuffle": true,
      "drop_last": true
    },
    "dataloader": {
      "num_worker": 16,
      "pin_memory": true
    },
    "tracker": [
      "tensorboard"
    ],
  }
}