Radio journalist recovers lost voice with artificial intelligence

Jamie Dupree has been described as the best-connected man in Washington. As the Washington correspondent for Cox Media Group, he has broadcast for various radio stations and speech is, obviously, central to his job.

 

In 2016 however, Dupree noticed his voice was beginning to fade and grow weak. He was soon was diagnosed with Tongue Protrusion Dystonia, a rare neurological condition which causes people to lose control over their tongues, making speech almost impossible.

He was given Botox injections to try and help but to this day doctors have been unable to find a medical solution. The Cox Group was committed to finding a solution so that Jamie, a key member of their team, could continue to work.

 

Enter CereProc, a Scottish tech company which built a new voice For Dupree using artificial intelligence.

 

Graham Leary, Head of Professional Services at CereProc, said;

“Even though [Jamie Dupree] can write blogs and articles, they wanted him to be able to speak and do daily addresses on radio because he is so well connected and knows all the politicians. So, they approached us and asked if we could do something because we did a similar project for an American film critic Roger Ebert who lost his voice after a thyroid complication.”

 

CereProc decided to make the most of Jamie’s 20+ years as a radio journalist and used archive recordings of his voice from over a short period of time (because our voices change over time as we age).

 

Graham Leary continues, “we had maybe six hours of audio in total and we did two builds of voice for him. With the very first build Jamie started to do his radio addresses again.”

 

In the second build, CereProc built a parametric voice which provides an entirely computer-generated representation of Jamie’s voice.

 

Explaining the process, Graham says “We take 6 to 8 hours of his audio and build a computer model that predicts how he will sound when he says a particular sentence”.

 

“We have a voice cloning system on our website, so people can record themselves and it takes about an hour and a half to two hours, but Jamie’s situation was much better because he has more data in his archive.”

 

It took CereProc about two months to re-create Dupree's voice, and Jamie is now able to continue working, doing his daily addresses. Cox Media was very happy to get their employee back doing his job and in addition to being able to produce broadcasts, he can use this technology for interviewing someone and for personal communication as well.

 

To speed up the process and make it more affordable, CereProc started developing its own neural networks in 2006. Graham said, “artificial intelligence systems can generate a voice in just a few days for £500, compared to building a voice from archive materials which can require 30 hours of recordings and is much more expensive”.

 

Jamie is broadcasting his voice again and it is all computer generated based on his original audio recordings. A result, Graham tells us, which has met with all round enthusiasm.

 

“I got an email from him last week saying that the reaction has been almost entirely positive.”

Views: 310

Reply to This

© 2023   Created by Gordon White.   Powered by

Badges  |  Report an Issue  |  Terms of Service