Google’s text-to-speech system has been impressive for a while now. That’s what you hear every time you talk Google Assistant. The once robotic voice now sounds remarkably human. But Google isn’t done improving it. A new system called Tacotron 2 is nearly indistinguishable according to Google.
A research paper published by Google shows a few examples of the Tacotron 2 system. This system trains neural networks to generate speech that sounds almost identical to a real human reading from a script. Google included a few examples to show how similar the AI and a human sound.
One of these audio clips is AI, the other is a human reading from text.
Pretty good, huh? Right now, the system is only trained to mimic one female voice. Google would need to train the system again for a different voice. This is very impressive stuff and should make it’s way to Goole Assistant in the future.