Google AI can replace you in speaking all kinds of foreign languages, with your own voice!

Before learning about Google's new AI technology, listen to the results of the research project first.

This is a sentence in Spanish.

Still the sentence, translated into English with Google's regular system. Other translation systems will have similar results, just different voices.

Translated into English, this sentence means: 'I wonder how I feel, and that's when I started crying'.

This is the original sentence translated into English, through the new automated translation system. The content of the sentence was different now: "Larry asked me how I felt, and that was when I started crying."

The results are not perfect, but you can see the intonation and the timbre in the sentence is much retained. This is Google's new instant translation system , which can allow users to retain the same voice.

This is the difference between the two old and new voice translation systems . Before this time, all sentences were converted into text, translated into the required language and re-read by mechanical voice. This causes intonation to be lost during translation.

The new system, given the name Translatotron , consists of 3 components; All three will focus on spectrum images - images that represent a certain spectrum, then perform three different tasks.

The first part will use neural networks that have been trained to analyze audio spectrum, to create an audio spectrum of the output language.

The second part will convert that spectrum into an audible sound wave.

The third part will separate the features in the original voice, to attach to the final sound clip.

The new way not only retains intonation (in many cases also implies) the speaker in the final product, on lyus theory, it also reduces errors when cutting unnecessary steps in the system Old translation system.

Translatotron is just a sample product that proves the system to work. In the process of testing, the researchers only focused on translating Spanish into English, 'only' could have needed huge amounts of AI training data. But it opened a narrow door to show us what the future might be like.