Google presented a program that allows simultaneous translation of human voice messages from one language to another – Translatotron. Until now, the old Google Translate first translated the spoken words into text and only then converted the text into a voice in the language chosen by the user.
With the new program you can do it directly and almost immediately, according to Gizmodo. Another feature of the new service will be the function of imitation – the program will be able to imitate the voice of the author of the message and even the tempo of his speech.
Winners of the annual Google Play prizes have been announced
Translatotron uses a "sequence-sequence" network model that accepts speech input, processes it as a spectrogram (visual representation of frequencies) and generates a new spectrogram in the target language.
As a result, the new program, as reported by Google, has an additional number of advantages over the existing methods. It will work faster, the number of errors reduced considerably. Processing of spoken words that do not need to be translated (for example, proper names) will improve.
According to Google, there are currently several text-to-speech applications, including Google Translate, SayHi, Microsoft Translator, iTranslate and TripLingo. However, none of them uses the user's real voice, so the company decided to correct this defect.