Google's new Translatotron AI can do speech-to-speech translation

Google has built a powerful new artificially intelligent system that is capable of translating the spoken word from one language to the other, without the need for text.

Called Translatotron, the new system is based on a sequence-to-sequence network that translates speech using source spectrograms as input and generates spectrograms of the translated content in the target language.

The highlight of the proposed system is that it is able to retain the vocal characteristics of the original speaker in the translated speech using speaker encoder network, making it sound more natural.