No idea of this is even vaguely in the right direction, but this comes to mind: Unsupervised speech-to-speech translation from monolingual data

https://research.google/blog/unsupervised-speech-to-speech-t...