xxx_cleaners about dl-art-school HOT 1 CLOSED

hi, you are talking about the text_cleaners that are used in paired_voice_audio_dataset.py, right?

# https://github.com/152334H/DL-Art-School/blob/master/codes/data/audio/paired_voice_audio_dataset.py#L157
        self.text_cleaners = hparams.text_cleaners

(note that, although the CharacterTokenizer is hard-coded to use "english_cleaners", the GPT model does not use a CharacterTokenizer -- it uses BPE.)

I am not sure whether this will work, but have you tried adding a text_cleaners parameter to the val/train dataset configs? e.g.

train:
  text_cleaners: ["basic_cleaners"]

The default cleaners are defined in the tacotron2 folder. They comment:

Cleaners are transformations that run over the input text at both training and eval time.

Cleaners can be selected by passing a comma-delimited list of cleaner names as the "cleaners" hyperparameter. Some cleaners are English-specific. You'll typically want to use:

1. "english_cleaners" for English text
2. "transliteration_cleaners" for non-English text that can be transliterated to ASCII using the Unidecode library (https://pypi.python.org/pypi/Unidecode)
3. "basic_cleaners" if you do not want to transliterate (in this case, you should also update the symbols in symbols.py to match your data).

If that advice is not helpful, then I must apologise, because I have little-to-no experience with working with other languages in TTS. I assume @athu16 did not do anything complicated to improve his setup, either.

from dl-art-school.