Closed avacaondata closed 2 years ago
Pinging @anton-l for knowledge :)
Hi @alexvaca0, thanks for your interest! We're currently working on implementing FastSpeech2 natively: https://github.com/huggingface/transformers/pull/15773 And looking into FastPitch as well: https://github.com/huggingface/transformers/issues/16349
Feel free to subscribe to those PRs/issues to follow the progress :)
Hey, it would be very cool if there were classes already developed for carrying out Text2Speech tasks. We do have many classes capable of doing Speech2Text, but for the opposite task there are no transformer classes available, although there are starting to be models in the hub: https://huggingface.co/facebook/tts_transformer-es-css10
It would be great if instead of using fairseq for loading those models, we could load them and fine-tune them further with Transformers.