The DeepMind AI Can Say Sentences and Play Piano Without a Sound Library

By: Joey Haar - Sep 12, 2016
References: deepmind & theverge
As algorithmic voices become a more common part of daily life, with Siri, Alexa and Cortana feeling as familiar as friends, the DeepMind AI's WaveNet program is focusing on creating an authentic sounding voice that is completely original. In other words, while Siri, Alexa, Cortana and others rely on processing recordings of a real human's voice, DeepMind AI's WaveNet creates an original voice by learning from a host of recordings.

The difference between WaveNet and other common systems is that WaveNet doesn't just put together sounds from a database. Rather, it learns speech sounds and puts those sounds together on its own. Because of this learning process, the DeepMind AI program can be used for anything that involves sound, including piano music (a sampling of which Google included on its blog.)