egs/tts/VALLE/exp_config.json

{
    "base_config": "config/valle.json",
    "model_type": "VALLE",
    "dataset": [
        "libritts"
    ],
    "dataset_path": {
        "libritts": "[LibriTTS dataset path]"
    },
    "preprocess": {
        "extract_phone": true,
        "phone_extractor": "espeak", // "espeak, pypinyin, pypinyin_initials_finals, lexicon (only for language=en-us right now)"
        "extract_acoustic_token": true,
        "use_phone": true,
        "use_acoustic_token": true,
        "processed_dir": "Amphion/data/",
        "sample_rate": 24000, // "Audio sampling rate." 
        "codec_hop_size": "320", // "Audio codec hop size." 
        "valid_file": "test.json", 
    },
    "model": {
        "prefix_mode": 1, // "The mode for how to prefix VALL-E NAR Decoder, 0: no prefix, 1: 0 to random, 2: random to random, 4: chunk of pre or post utterance.",
    },
    "log_dir": "Amphion/ckpts/tts/valle",
    "train": {
        "batch_size": 4,
        "train_stage": 1, // 0: train all modules, For VALL_E, support 1: AR Decoder 2: NAR Decoder(s)
        "max_epoch": 20, // "Number of epochs to train."
        "use_dynamic_batchsize": true, // If use dynamic batch size
        "max_tokens": 4000, // If use dynamic batch size
        "max_sentences": 10 // If use dynamic batch size
    }
}