News
News Categories

Google's new AI is now capable of perfectly imitating human speech

By John Law & James Lu - on 28 Dec 2017, 10:45am

Google's new AI is now capable of perfectly imitating human speech

Google has created a new text-to-speech AI called Tacotron 2, which uses deep neural networks to perfectly imitate human speech. Google published a research paper about the system, which is able to read text in a generated female voice that is indistinguishable from an actual human reading the same text.

The system is Google’s second official generation of the technology, which consists of two deep neural networks. The first network translates the text into a spectrogram, a visual way to represent audio frequencies over time. That spectrogram is then fed into WaveNet, a system from Alphabet’s AI research lab DeepMind, which reads the chart and generates the corresponding audio elements accordingly.

You can check out audio samples of Tacotron 2 and compare it to an actual human speaking the same text here.

Google via Quartz.