Training using "ipa" tokenizers #467

New Issue

bahm9919 · 2024-01-19T01:20:14Z

bahm9919 commented

2024-01-19 01:20:14 +00:00

Guys has anyone tried to use ipa tokenizers for training? I can't, get the error "expected string or bytes-like object".

MisterCapi commented

2024-01-21 21:00:48 +00:00

You would have to provide more details to help you in this specific case, but if you are going into phonemes I'd suggest creating your own phonemizer using kaikki's dictionary and falling back onto the default what this repo uses [espeak NG]. Espeak doesn't really handle some words well in my experience.

You would have to provide more details to help you in this specific case, but if you are going into phonemes I'd suggest creating your own phonemizer using [kaikki's dictionary](https://kaikki.org/dictionary/English/index.html) and falling back onto the default what this repo uses [espeak NG]. Espeak doesn't really handle some words well in my experience.

Sign in to join this conversation.