train_vits.py

import os

from trainer import Trainer, TrainerArgs

from TTS.tts.configs.shared_configs import BaseDatasetConfig,BaseAudioConfig,CharactersConfig
from TTS.tts.configs.vits_config import VitsConfig
from TTS.tts.datasets import load_tts_samples
from TTS.tts.models.vits import Vits, VitsAudioConfig
from TTS.tts.utils.text.tokenizer import TTSTokenizer
from TTS.utils.audio import AudioProcessor

output_path = os.path.dirname(os.path.abspath(__file__))
RESTORE_PATH = '/home/azureuser/BanglaTTS/nctb-vits-single-female-1/checkpoint.pth'
SPEAKER_ID = 9
SPEAKER_GENDER = 'male'
meta_file = f"/home/azureuser/BanglaTTS/nctb-audiobook-no-numbers/{SPEAKER_GENDER}/SP_{SPEAKER_ID}/metadata.txt"
root_path = f"/home/azureuser/BanglaTTS/nctb-audiobook-no-numbers/{SPEAKER_GENDER}/SP_{SPEAKER_ID}"

def formatter(root_path, meta_file, **kwargs):  # pylint: disable=unused-argument
    """Normalizes the LJSpeech meta data file to TTS format
    https://keithito.com/LJ-Speech-Dataset/"""
    txt_file = meta_file
    items = []
    speaker_name = f"nctb_{SPEAKER_GENDER}_{SPEAKER_ID}"
    with open(txt_file, "r", encoding="utf-8") as ttf:
        for line in ttf:
            cols = line.split("|")
            wav_file = os.path.join(root_path,'audio', cols[0])
            try:
                text = cols[1]
            except:
                print("not found")

            items.append({"text": text, "audio_file": wav_file, "speaker_name": speaker_name, "root_path": root_path})
    return items


dataset_config = BaseDatasetConfig(
     meta_file_train=meta_file, path=os.path.join(root_path, "")
)

characters_config = CharactersConfig(
    pad = '<PAD>',
    eos = '<EOS>', #'<EOS>', #'।',
    bos = '<BOS>',# None,
    blank = '<BLNK>',
    phonemes = None,
    characters =  "abcdefghijklmnopqrstuvwxyz0123456789+=/*√তট৫ভিঐঋখঊড়ইজমএেঘঙসীঢ়হঞ‘ঈকণ৬ঁৗশঢঠ\u200c১্২৮দৃঔগও—ছউংবৈঝাযফ\u200dচরষঅৌৎথড়৪ধ০ুূ৩আঃপয়’'”^নলো_…ৰ",
    #characters =  "তট৫ভিঐঋখঊড়ইজমএেঘঙসীঢ়হঞ‘ঈকণ৬ঁৗশঢঠ\u200c১্২৮দৃঔগও—ছউংবৈঝাযফ\u200dচরষঅৌৎথড়৪ধ০ুূ৩আঃপয়’নলোˌamɾʃˈonbŋlitjʰɔdkpeɟːfɡuhrʈæsʒɖwəc",
    punctuations = "-–:;!,|.?॥। “",
)

#ণ´0ুয)wCছ=ক'স_{rMথd“ো+W।চঋ৷ঔ…’Eৰওঢxoঝূৎ5iটআইSyAc—ড√ল8ঁিk়াYVz‍ফLbD-শlপ য়–গ(রঐ্ঊ‘অ‌Gঈষgভ!:n;ীO?vড়aq/tRঘবএঠpধ
#ংখJঙঢ়]ৃউNহত,”নৗIfBৈmP॥sueঃৌhFমজদঞT.*েHj[

audio_config = VitsAudioConfig(
    sample_rate=16000, win_length=1024, hop_length=256, num_mels=80, mel_fmin=0, mel_fmax=None
)

# VitsConfig: all model related values for training, validating and testing.

config = VitsConfig(
    audio=audio_config,
    run_name="vits-ft-nctb",
    batch_size=48,
    eval_batch_size=8,
    batch_group_size=5,
    num_loader_workers=8,
    num_eval_loader_workers=4,
    run_eval=True,
    test_delay_epochs=-1,
    epochs=35,  # testing
#     phonemizer="bn_phonemizer",# multi_phonemizer
    text_cleaner='multilingual_cleaners',#'multilingual_cleaners', #"collapse_whitespace" phoneme_cleaners multilingual_cleaners
    use_phonemes=False,
#     phoneme_language="bn",

#     phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
    compute_input_seq_cache=True,
    add_blank=True,
    use_language_weighted_sampler = True,
    print_step=500,
    print_eval=False,
    mixed_precision=True,
    output_path=output_path,
    datasets=[dataset_config],
    characters = characters_config,
    save_step=1000,
    cudnn_benchmark=True,
    # dashboard_logger = 'wandb',
    test_sentences = [
        ["আমরা বাংলায় ওয়েব ডেভেলপমেন্ট নিয়ে কাজ করতে গিয়ে প্রথম যে সমস্যাটার মুখোমুখি হই, সেটা হলো, বাংলা ডেমো টেক্সট"],
        ["আমি বাঙালি ভাষায় কথা বলতে পারি।"],
        ["আমরা প্রকৃতি কে ভালোবাসি।"],
        ["আপনি কেমন আছেন?"],
    ]
)

# INITIALIZE THE AUDIO PROCESSOR
# Audio processor is used for feature extraction and audio I/O.
# It mainly serves to the dataloader and the training loggers.
ap = AudioProcessor.init_from_config(config)

# INITIALIZE THE TOKENIZER
# Tokenizer is used to convert text to sequences of token IDs.
# config is updated with the default characters if not defined in the config.
tokenizer, config = TTSTokenizer.init_from_config(config)

# LOAD DATA SAMPLES
# Each sample is a list of ```[text, audio_file_path, speaker_name]```
# You can define your custom sample loader returning the list of samples.
# Or define your custom formatter and pass it to the `load_tts_samples`.
# Check `TTS.tts.datasets.load_tts_samples` for more details.
train_samples, eval_samples = load_tts_samples(
    dataset_config,
    formatter=formatter, 
    eval_split=True,
    eval_split_max_size=config.eval_split_max_size,
    eval_split_size=config.eval_split_size,
)

# init model
model = Vits(config, ap, tokenizer, speaker_manager=None)

# init the trainer and 🚀
trainer = Trainer(
    TrainerArgs(restore_path = RESTORE_PATH), 
    config, 
    output_path, 
    model=model, 
    train_samples=train_samples, 
    eval_samples=eval_samples,
)
trainer.fit()