train_vits.py

import os

from trainer import Trainer, TrainerArgs

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

output_path = os.path.dirname(os.path.abspath(__file__))
# dataset_config = BaseDatasetConfig(
#     formatter="vctk", meta_file_train="", language="en-us", path=os.path.join(output_path, "../VCTK/")
# )

CONTINUE_PATH=None
RESTORE_PATH=None
START_WITH_EVAL=True
GRAD_ACUMM_STEPS=1

meta_file = '/home/ubuntu/nctb-cropped/metadata.txt'
root_path = '/home/ubuntu/nctb-cropped/'

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 = []
    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])
            speaker_name = cols[0].split('_')[-1].split('.')[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, ""), language="bn"
)


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


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

vitsArgs = VitsArgs(
    use_speaker_embedding=True,
)

config = VitsConfig(
    model_args=vitsArgs,
    audio=audio_config,
    run_name="vits_nctb",
    batch_size=64,
    eval_batch_size=8,
    batch_group_size=5,
    num_loader_workers=4,
    num_eval_loader_workers=4,
    run_eval=True,
    test_delay_epochs=-1,
    epochs=1000,
    # text_cleaner="english_cleaners",
    text_cleaner='phoneme_cleaners',
    use_phonemes=True,
    phoneme_language="bn",
    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
    compute_input_seq_cache=True,
    print_step=25,
    print_eval=False,
    mixed_precision=True,
    max_text_len=325,  # change this if you have a larger VRAM than 16GB
    output_path=output_path,
    datasets=[dataset_config],
    characters=characters_config,
    save_step=5000,
    cudnn_benchmark=True,
    test_sentences = [
        'আমার   সোনার বাংলা, আমি তোমায় ভালোবাসি।',
        'চিরদিন   তোমার আকাশ, তোমার বাতাস, আমার প্রাণে বাজায় বাঁশি', 
        'ও মা,   ফাগুনে তোর আমের বনে ঘ্রাণে পাগল করে,মরি হায়, হায় রে।'
    ]
)

# 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 speaker manager for multi-speaker training
# it maps speaker-id to speaker-name in the model and data-loader
speaker_manager = SpeakerManager()
speaker_manager.set_ids_from_data(train_samples + eval_samples, parse_key="speaker_name")
config.model_args.num_speakers = speaker_manager.num_speakers

# init model
model = Vits(config, ap, tokenizer, speaker_manager)

# init the trainer and 🚀
trainer = Trainer(
    TrainerArgs(
        continue_path=CONTINUE_PATH,
        restore_path=RESTORE_PATH,
        skip_train_epoch=False,
        start_with_eval=START_WITH_EVAL,
        grad_accum_steps=GRAD_ACUMM_STEPS,
    ),
    config,
    output_path,
    model=model,
    train_samples=train_samples,
    eval_samples=eval_samples,
)
trainer.fit()