According to Engadget , Translatotron skips the usual step of translating speech into text and turning into speech, which uses end-to-end technology to directly translate the speaker’s voice into another language. The company hopes the technology will open up future developments in language translation.
Gooogle said, Translatotron uses the network model in order to take voice input, process it in the form of a spectrum – visualize the frequencies – and create a new spectrum in the target language. The result is a much faster translation with less data lost during the migration process.
In the short term, a translated test speech has quality robot steam output, but still effectively maintains some elements of the voice. Anyway, this is just the beginning, so everything promises to improve even more in the future.
In addition, Google has refined its translations in recent months. Last year, the company introduced new features for Google Translate so it could speak many other languages in regional-based pronunciation and add multiple languages to its real-time translation feature.