A futuristic institute aims to preserve (and recreate) people's voices using artificial intelligence.
The researchers involved in the "voice bank" initiative hope to be able to change the lives of all people who find themselves losing their voice following illness or accidents. Losing your voice is not exactly the same as having a temporary decrease in sound or hoarseness. The loss of voice is partly the loss of identity.
It all started from a collaboration between Northeastern University of Boston and the Vowels D.. Offering those who lose the ability to speak a way to maintain a sense of identity after throat cancer or a degenerative disease is important. This project will allow them to “speak” using a synthetic form very similar to their own voice.
It is the first center of this type, and will be led by prof. Rupal Patel, founder and CEO of VocaliD.
The company already offers such a service for individuals, Patel says, but many people don't have access to high-quality equipment to properly sample their voices. Creating a “voice bank” allows future patients to preserve their voice while they still have the chance.
“Patients often come to us at the last minute,” He says. “They don't have enough time to preserve their voice because they are at the mercy of illness, operations and more. And it's very frustrating."
The voice bank
The result was the collaboration with Northeastern to spread the technology to the public. The heart of the project is the Voice Preservation Clinic, a center where patients at risk of losing their voice can register it to protect it.
The approach is much more complicated than those based on traditional speech synthesis: a machine learning algorithm reconstructs ligatures, phonemes and even rhythm in relation to breathing. In a special booth, small stories, poems, or conversations on different themes are recorded. Someone even joked about how to lose your voice, a self-deprecating way of dealing with great fear.
“What you need is about two or three hours of speaking. From these recordings our artificial intelligence is able to generate a speech engine that sounds the same. It's like giving the voice back to those who have lost it."
The regenerated voice can be used as speech synthesis to speak by converting written texts, and can be integrated tomorrow with software that generates words, images or actions from mental paths.
A very rapid improvement
Technology, Patel says, is improving rapidly, and will soon be able to reproduce speech that is indistinguishable from human speech.
The team will also be able to "age" the voice to adapt it to the person's age, while it is not yet possible to "grow" a child's voice into an adult.