config/fs2.json

{
    "base_config": "config/tts.json",
    "model_type": "FastSpeech2",
    "task_type": "tts",
    "dataset": ["LJSpeech"], 
    "preprocess": {
      // acoustic features
      "extract_audio": true,
      "extract_mel": true,
      "mel_extract_mode": "taco",
      "mel_min_max_norm": false,
      "extract_pitch": true,
      "extract_uv": false,
      "pitch_extractor": "dio",
      "extract_energy": true,
      "energy_extract_mode": "from_tacotron_stft",
      "extract_duration": true,
      "use_phone": true,
      "pitch_norm": true,
      "energy_norm": true,
      "pitch_remove_outlier": true,
      "energy_remove_outlier": true,

      // Default config 
      "n_mel": 80,
      "win_size": 1024,  // todo
      "hop_size": 256,
      "sample_rate": 22050,
      "n_fft": 1024, // todo
      "fmin": 0,
      "fmax": 8000, // todo
      "raw_data": "raw_data",
      "text_cleaners": ["english_cleaners"],
      "f0_min": 71,    // ~C2
      "f0_max": 800, //1100,    // ~C6(1100), ~G5(800)
      "pitch_bin": 256,
      "pitch_max": 1100.0,
      "pitch_min": 50.0,
      "is_label": true,
      "is_mu_law": true,
      "bits": 8,

      "mel_min_max_stats_dir": "mel_min_max_stats",
      "whisper_dir": "whisper",
      "content_vector_dir": "content_vector",
      "wenet_dir": "wenet",
      "mert_dir": "mert",
      "spk2id":"spk2id.json",
      "utt2spk":"utt2spk",

      // Features used for model training
      "use_mel": true,
      "use_min_max_norm_mel": false,
      "use_frame_pitch": false,
      "use_frame_energy": false,
      "use_phone_pitch": true,
      "use_phone_energy": true,
      "use_log_scale_pitch": false,
      "use_log_scale_energy": false,
      "use_spkid": false,
      "align_mel_duration": true,
      "text_cleaners": ["english_cleaners"]
      },
    "model": {
      // Settings for transformer
      "transformer": {
        "encoder_layer": 4,
        "encoder_head": 2,
        "encoder_hidden": 256,
        "decoder_layer": 6,
        "decoder_head": 2,
        "decoder_hidden": 256,
        "conv_filter_size": 1024,
        "conv_kernel_size": [9, 1],
        "encoder_dropout": 0.2,
        "decoder_dropout": 0.2
      },

      // Settings for variance_predictor
      "variance_predictor":{
        "filter_size": 256,
        "kernel_size": 3,
        "dropout": 0.5
      },
    "variance_embedding":{
        "pitch_quantization": "linear", // support 'linear' or 'log', 'log' is allowed only if the pitch values are not normalized during preprocessing
        "energy_quantization": "linear", // support 'linear' or 'log', 'log' is allowed only if the energy values are not normalized during preprocessing
        "n_bins": 256
      },
    "max_seq_len": 1000
    },
    "train":{
      "batch_size": 16,
      "sort_sample": true,
      "drop_last": true,
      "group_size": 4,
      "grad_clip_thresh": 1.0,
      "dataloader": {
        "num_worker": 8,
        "pin_memory": true
      },
      "lr_scheduler":{
        "num_warmup": 4000
      },
      // LR Scheduler
      "scheduler": "NoamLR",
      // Optimizer
      "optimizer": "Adam",
      "adam": {
        "lr": 0.0625,
        "betas": [0.9, 0.98],
        "eps": 0.000000001,
        "weight_decay": 0.0
      },
    }

}