Hiragana Katakana and Kanji Support

[641i130]

I just discovered this project, really well done! Also nice music tastes 🤝

How much work would it take to support hiragana/katakana and kanji? Would it need to be converted to phonemes?

I'm new to this space, but I'm trying to write a language learning tool that takes lyrics to a song and shows the meaning of each word literally for vocabulary practice.

I'd be willing to write this if you could possibly point me towards the files to modify. I'm guessing its also possible the model in use doesn't support the characters, but I'm not sure.

[NextFire]

Hi! Thank you for your interest in this project!

Yohane is more or less a wrapper for the torchaudio "Forced alignment for multilingual data" tutorial, which relies on the MMS_FA model.

I'm not sure what you want to achieve, but you can take a look at the "Chinese" section of the aforementioned tutorial, which states that:

In order to obtain word level alignments, you need to first tokenize the transcripts at the word level using a word tokenizer [...]. However this is not needed if you only want character-level alignments.

Depending of I designed well or not the library, you may not need to directly edit this project. Instead, you can programmatically use or extend the Yohane pipeline class (see the "Run" section in the notebook).