Text-to-text and Speech-to-speech in a single model inspired by the Babel Fish from the The Hitchhiker's Guide to the Galaxy