This is called end-to-end speech translation, and has been around since 2017. Here's an article from 2019: https://www.technologyreview.com/2019/05/20/103054/google-ai...