There is also many data from audiobooks in many languages that are easy to scrap and align using a basic model that has been updated for each language, or using

There is also many data from audiobooks in many languages that are easy to scrap and align using a basic model that has been updated for each language, or using Youtube videos with subtitles that are almost aligned for a first version of the model, then realigning