Hebrew is fundamentally a hard language to work in the field of Natural language processing and it is also one of the underrepresented language in the field of Speech-Speech and Text-to-Speech models. Mainly boils down to limited availability of data. To explore Speech-Speech
(Voice Cloning), I used
Dataset to fine-tune Fish-speech 1.5 on roughly 2.5 hours of Hebrew audio on their Gold-standard subset.
I have also fixed a few bugs on Fish's fine-tuning code and created a
pull-request
