The Translatotron 2 model consists of a source speech encoder, a target phoneme decoder, and a target mel-spectrogram synthesiser.

The post Google Releases New Version of Translatotron: Its End-to-end Speech Translation Model appeared first on Analytics India Magazine.