The developers at DeepMind has taught artificial intelligence to emulate the human voice
The company’s specialists DeepMind bought by Google in 2014 for $ 400 million and specializes in the development of artificial intelligence, created a system of WaveNet, which can imitate the voice of man.
In programs that generate the voice, taken two approaches. In one use a huge amount of ready-made snippets of human speech, but to manipulate the sounds and intonations difficult. In another the sounds are synthesized. They are easy to modify, but it has a distinctive «mechanical» sound.
To teach the computer to emulate human speech, the developers have processed recorded samples using the neutron network, which was able to extract the information responsible for the similarity with natural speech, and apply it to synthesized sounds. Interestingly, samples can be not only a speech system with the same success handled the samples of the piano sound, highlighting information that allows to simulate the sound of the instrument.