espnet3.systems.f5tts.src.preprocessor.F5PinyinPreprocessor

source

class espnet3.systems.f5tts.src.preprocessor.F5PinyinPreprocessor(vocab_file: str, text_name: str = 'text')

Bases: object

Tokenize text with F5’s zh+en pinyin scheme using F5’s fixed vocab.

Configure the preprocessor.

• Parameters:
  • vocab_file – F5’s vocab.txt (e.g. Emilia_ZH_EN_pinyin/vocab.txt); line number = token id, space at index 0.
  • text_name – Sample dict key holding the raw transcript.

Example

preprocessor:
  _target_: espnet3.systems.f5tts.src.preprocessor.F5PinyinPreprocessor
  vocab_file: /path/to/Emilia_ZH_EN_pinyin/vocab.txt

NOTE

The vocabulary is read once at construction, so vocab_file must be the same one the checkpoint was trained with: it fixes the token ids, and a different file silently remaps them.

property vocab_size : int

Return the number of tokens in F5’s fixed vocabulary.

• Returns: Line count of vocab_file.

NOTE

This is the value to use as the model’s idim when training against F5’s fixed vocabulary instead of an ESPnet token list.