The startup Lyrebird announced in a press release on their website that they would be releasing an API to accurately replicate human voices. The new company is a spinoff founded by researchers of the MILA lab at the University of Montréal.
The new API will be able to accurately reproduce someone's voice by creating a unique voice identifier key from audio fragments. Only 1 minute of audio should already be enough to make the end result quite life-like, which is why this technology is so ground breaking.
The software doesn't only reproduce the voice, but also allows the produced voice to have different kind of pitches and intonations, without losing its resemblance to the original. For those who are curious what the result sounds like, there are plenty of fragments on the demo page of Lyrebird's website, such as a "discussion" between the voices of well-known US politicians about their software suite.
Of course, a groundbreaking development like this one carry some concerns with them. Being able to replicate anyone's voice with free software tools allows one to impersonate anyone they have a voice recording of. Especially in court cases, a voice recording will have to be dismissed as evidence, as now it is virtually impossible to make the distinction between a real recording and a fabricated one. The Lyrebird team discusses this topic as well on their website, saying they wish to publicly release the API to make sure everyone has equal access to it, making sure the existence of it is well-known, so that society is alert to its repercussions.
The new technology sure brings a lot of exciting new possibilities: replica's of actors' voices in movies, giving life-like voices to AI robots, etc..
However, as with all new technologies, legislation will have to catch up to dictate how society will accept this new change into their daily lives. I, for one, am curious what the future will bring when the Lyrebird API gets actually released to the public, and other companies can adopt the new technology into their services.