After losing the ability to speak due to a rare neurological condition, US radio journalist Jamie Dupree will return to the air by using the speech synthesis technology.

By Daniel Herborn

Posted on June 15, 2018

The 54-year-old Dupree has been a radio journalist for a number of stations in the Cox Media Group stable, covering US politics. He has been broadcasting since 1983.

New technology created specifically for him by Scottish company CereProc will allow him to return to the airwaves from Monday 25 June 2018. Dupree will write a script for his broadcasts and then use free text-to-speech software Balanolka to turn it into an audio recording which replicates his voice.

“Jamie Dupree 2.0 is here – and I couldn’t be more excited about it” he wrote on his blog.

The technology will draw on the extensive radio archives built up from Dupree’s previous broadcast work on radio. CereProc built a neural network that would predict Dupree’s speech patterns from these recordings.

He will also have the ability to isolate and slow down certain consonants or vowels and to swap words to produce natural-sounding audio.

“It is me, there is no doubt about that,” Dupree wrote.

“Yes, it is slightly robotic, but no-one was promising me that it was going to be perfect.”

Dupree told the BBC the development of the technology “saved my job and saved my family from a terrible financial unknown.”

“There is not much of a market for radio reporters who can’t talk,” he added.

Dupree lost his voice two years ago

During the buildup to the 2016 US presidential election, Dupree‘s voice became wheezy and unrecognisable before he lost the ability to speak completely. He underwent extensive tests to determine the problem and was given botox injections in the mouth in an attempt to relax his tongue.

Eventually doctors discovered he had been affected by a neurologic condition called muscle tension dystonia. The condition is so rare that no doctors specialise in treating it. The cause of the disorder remains unknown.

Dupree continued to work as a journalist, interviewing politicians by writing questions on an eWriter tablet. He also records responses to questions asked by other journalists in media scrums and turns these into written stories. Since being forced off air he has become a prolific commentator on social media.

His efforts to continue his career in trying circumstances were recognised by Republican congresswoman Ileana Ros-Lehtinen who paid tribute to Dupree in a speech in the US House of Representatives in December 2017.

Not only did the congresswoman’s tribute raise Dupree’s flagging spirits, it ignited interest in his case across news organisations and the medical community. Eventually he was referred to Dr. Hyder Jinnah at the Emory University Brain Health Centre who made the diagnosis in March 2018. Dupree will return to Emory University for further treatment in August this year.

A new use for AI

CereProc has been involved in developing neural networks since 2006. Its artificial intelligence system divides each word into 100 fragments and repeats this process until it builds up an understanding of how that speaker strings certain words together.

Chris Pidcock, co-founder of CereProc, said there was significant potential for artificial intelligence to be applied in this area. “All techniques work quite well on small constrained problems, and learning to model speech is something deep neural nets can do really well.”

In this instance, CereProc had extensive examples of Dupree’s voice to work with but it says it can build a new voice for a speaker from as little as 620 sentences.

Rich Jones, a colleague of Dupree’s at WOKV radio in Jacksonville, Florida says he had “an emotional reaction” when he heard Dupree’s new voice for the first time.

Dupree himself is exuberant about his return to radio. “Writing for my blog, sending out tweets and doing Facebook is great – but there is nothing like cranking out a 20-second story jammed with a couple of sound bites to make the top of the hour newscast,” he wrote.