Google launched a new text-to-speech system, which he calls Tacotron 2. The system works with stunning accuracy, generating an artificial voice indistinguishable from the voice of a real person.

This is not an exaggeration: Tacotron 2 is the second generation of technology and consists of two neural networks. The first of them converts the text into a special spectrogram (you'll see it in the figure below), and the second, WaveNet, "reads" this diagram and interprets it, creating a voice indistinguishable from the human.


A spectrogram of the exclamation "Whoah". Image: Google
The new system does not sound like "robot voice" and can even correctly pronounce words depending on the semantics.

Now the system learns all the nuances of the English language and can speak only one female voice. The system can not only read the text, but also allocate certain places in the speech: if a word is selected in lower case letters, it will accentuate it.

![ai.png](https://steemitimages.com/DQmZcEGJeqhB2XJnheSJdT7mmsmSchfVCVUS3gYzymgxtU4/ai.png)