The neural network has learned to replicate the human voice

Scientists from Massachusetts have created a neural network that can duplicate and change people's voices, even in real time. So, on the main page, the startup Modulate.ai is advertised by Barack Obama, more precisely his voice, modeled by artificial intelligence.

Will Knight, a journalist from MIT Technology Review (journal owned by Massachusetts Institute of Technology), decided to try a new development. He decided to listen to how his voice sounded in the female version.

The observer noted that at the beginning of work, the voice generated by the neural network was shaking slightly. But the longer the communication lasted, the closer the sound became to the living human voice.

Modulate uses a Generation-Contest network (GAN) to capture and simulate the sound properties of a voice signal.

Mike Pappas, the head of Modulate.ai, claims that the startup was created with good intentions and does not set a goal to copy the voices of popular people and opinion leaders. To warn users about what the neural network says with them, Mike plans to insert watermarks into audio tracks generated by artificial intelligence.

Modulate is focused on online games, such as Fornite or Call of Duty, in which players can communicate with strangers through a microphone. According to the developers, the use of AI will help improve the gameplay and remain anonymous to those players who do not want to use their real voice, able to reveal their age or gender.

robo-hunter.com