Google has finally launched “Translatotron” which is a voice-to-voice translation tool that translates one language to another language without typing any text. After a number of years of research and development, this translation tool is made, but this is still in the development phase.
Most of the other translation tools first change the speech to a text format, and then this text is finally changed to a speech. But when it comes to Google’s Translatotron – New Speech Translation tool, it directly translates the speaker’s voice to the voice of another desired language.
Google Research’s GitHub Page says that this tool is an “attention-based sequence-to-sequence neural network which can directly translate speech from one language into speech in another language, without relying on an intermediate text representation”.
According to Google, this tool forms a spectrogram of the input voice or speech, and a new spectrogram of the output language is created. The final translated speech is a combination of a bit robotic and some elements of the speaker’s voice. A spectrogram is a visual representation of frequencies of a sound.
This tool is fast as compared to other tools in translating because of Google’s New Speech Translation tool which skips the step of translating the speaker’s voice to text and then text to voice of the selected language.
Google has tested the Translatotron on a number of other languages. You can check out all the sample on Google’s Research GitHub page. This tool is not perfect and a lot of development is to be made to make this tool produce a result which will be similar to the speaker’s voice and does not involve any robotic feature. The output result would be natural to hear.