Google introduced a program that allows simultaneous translations of human voice messages from one language to another – Translatotron. So far, the old Google Translator first translates the spoken words into text and only then translates the text into a voice of the selected language.
The new program will allow you to do it directly and almost instantly, according to Gizmodo. Another feature of the new service will be the imitation function – the program will be able to mimic the voice of the author of the message and even the rhythm of his speech.
The winners of the annual awards on Google Play have become famous
Translatotron uses a "sequence-sequence" networking model that accepts voice input, processes it as a spectrogram (visual frequency representation), and generates a new spectrogram in the target language.
As a result, as Google reports, the new program has a number of advantages over existing methods. It will work faster, the number of errors significantly decreases. The processing of spoken words that do not need to be translated (for example, proper names) will be improved.
At present, Google claims that there are several word-of-speech applications, including Google Translate, SayHi, Microsoft Translator, iTranslate and TripLingo. None of them, however, uses the true voice of the user, so the company decides to correct this defect.