Lyrebird was founded by three PhD students from the Montréal Institute for Learning Algorithms (MILA): Jose Sotelo, Kundan Kumar, and Alexandre de Brébisson. Their groundbreaking academic research applied deep learning, a type of artificial intelligence, to speech synthesis. Older methods of voice copying required hours of audio. Lyrebird’s approach can copy a voice with as little as one minute of recorded speech.
Any single human voice is very complex. But groups of voices share many characteristics. Lyrebird starts with a general voice model, trained on many speakers. Then, the idiosyncratic features of the ‘target’ voice – the voice to be copied – are identified and are used to adapt the general model. In this way, Lyrebird can create a ‘vocal avatar’ for any individual speaker using a very small amount of speech data.
What’s more, Lyrebird’s voices are capable of sounding emotional. This opens up a whole host of practical and creative opportunities: from personalized voice assistants to avatars for video games, or audiobooks read in the voice of a loved one. And there is great potential to use these voices in healthcare. Lyrebird wants to offer personalized assistive technology for people who can’t speak, and for visually-impaired people who use screen readers to have text read aloud to them.
One of Lyrebird’s proudest partnerships is with the ALS Association, on ‘Project Revoice’. Project Revoice is a non-profit initiative helping people with ALS to bank and digitally copy their voice. ALS (also known as Motor Neurone Disease / Lou Gehrig’s Disease) is a progressive neurodegenerative disease that affects nerve cells in the brain and spinal cord. Over time, muscles in the mouth, throat, and chest may be affected, which in turn is likely to affect an individual’s ability to speak. Losing the power of speech is typically a very frustrating and distressing effect of ALS; our voices are a big part of our personal identities. Lyrebird is passionate about applying their technology to help.
Project Revoice’s jumping-off point was the recreation of Pat Quinn’s voice. Pat Quinn is a co-founder of the successful Ice Bucket Challenge, who himself is living with ALS. Pat had not banked his voice before he lost the power of speech. Instead, Lyrebird used recordings from Pat’s many Ice Bucket Challenge interviews to make a copy of his voice. Project Revoice then linked Pat’s Lyrebird voice with assistive technology, enabling him to generate speech in his original, unique voice – rather than a generic one.
The most exciting part is: this is just the start. Lyrebird’s technology is still at a relatively early stage. Montréal has established itself as a center of AI research, attracting the best and brightest companies and researchers. The Lyrebird team is dedicated and diverse, assembled in Montréal from many different places around the world, with a common ambition for the future of speech synthesis technology.