What should we do for Homographs Disambiguation?

[cod3r0k]

Hi,
In most languages, there are words with different meanings depending on their position in the text. Maybe we need to perform some text preprocessing steps like NER (Named Entity Recognition), POS (Part of Speech) tagging, or similar techniques to identify these words. In Arabic or most Unicode languages, we must disambiguate such words, add diacritics, or use a better word form to handle stress, pronunciation, and other nuances.

What should we do in Coqui and VITS? Do you suggest any specific approach for this?
Also, when do you recommend applying these steps? Before feeding data to the model, or later?

For example, during training, would you suggest applying these transformations to the dataset itself (ensuring the dataset is well-cleaned)?
Or should we modify cleaners.py in Coqui to handle these tasks during training and inference?

[eginhard]

Yes, you can either do any text processing first on your side or integrate it into the Coqui pipeline with custom cleaners.