Posted: 1:00 pm Sunday, June 17th, 2018
By Jamie Dupree
Monday marks the start of a new effort to get my voice back on the radio for the first time in two years, by using a high tech solution, a computer generated voice, drawn from recordings of my old stories, as medical efforts to bring my voice back – to anything close to normal – have not been successful.
It was April 2016 when my voice began to falter, after I got sick on a family vacation; since then, my doctors have determined that I have a rare neurological disorder, in which the signals from the brain are getting mixed up somehow, causing my tongue to push out of my mouth when I speak – it’s known as ‘tongue protrusion dystonia.’
As it became obvious in the last year that my voice was not coming back, we searched for answers, and finally, high tech guru Mike Lupo at our Cox Media Group corporate headquarters contacted a company in Scotland, CereProc, which agreed to try to build what amounts to a Jamie Dupree voice app.
How does it work? How do I produce stories with it? Why is it even needed? Let’s take a look.
1. Let’s start with an explanation of what’s wrong. Over the past two years, there have been no answers in the search for my voice. What I have is a neurological disorder, for which there really aren’t specific treatments, known as tongue protrusion dystonia. When I try to talk, my tongue pops out of my mouth, my throat clenches, and it results in a strangled, unintelligible voice. I’ve been to Johns Hopkins, Georgetown, George Washington, the Cleveland Clinic, Emory University. The Mayo Clinic wouldn’t take my case. Many doctors have frankly admitted my problems were above their expertise. At Easter of 2017, the head of the voice center at the Cleveland Clinic correctly diagnosed my problems, but had no names to offer me in terms of treatment. I have been seeing a doctor outside of D.C. who agreed to try to decipher my case, but we really haven’t pushed any closer to a solution. In March, I saw Dr. Hyder Jinnah at the Emory University Brain Health Center in Atlanta – he gave me two Botox shots to my tongue in mid-May, to see if that would slow my tongue, and stop it from thrusting out of my mouth when I speak. It hasn’t really helped, so we will try again in August, with a little more Botox. You can hear from my doctor in this report that was done by CNN’s medical unit, thanks to producer Sandee LaMotte.
If you want the best explanation of the medical side of my voice problems, this is it https://t.co/d0aH21Za1b
— Jamie Dupree (@jamiedupree) June 16, 2018
2. Building a voice from the Dupree archives. Since my voice isn’t going to re-appear anytime soon, we started looking for high tech ways to get me back on the air. The first step was gathering years of recordings of my voice. I have shoe boxes filled with cassette tapes, reel-to-reel tapes, mini-discs, hard drives, and all sorts of different media storage devices from the first time I went on the radio in 1983, up through April of 2016 when my voice went out. But to build this voice, we focused on recordings from the last few years, which I had saved on our company computer system. Whether it was my reports from Capitol Hill, or from out on the campaign trail, I had hours and hours of material. But what the people at CereProc needed was audio that was only from me – so I spent several very late nights sorting through hundreds and hundreds of my stories to isolate those items which would help build a good voice. Going through all of that audio, it was like a trip down memory lane of what news stories that I had covered in the past few years, where I had been during my campaign coverage, what stories were big, and more. But that audio search was also a sobering personal reminder for me, that what was once normal – the mere act of speaking on the radio – was now impossible. Thus, the need for Jamie Dupree 2.0.
3. CereProc then goes to work. Once I handed off hundreds of audio files to the folks at CereProc in Scotland, all I could do was wait to see what they were going to be able to produce. “The voice was harder to build as the audio data used to build the voice was not recorded for the purpose of building a text-to-speech voice,” said Graham Leary, who was in charge of my voice development. “Normally we would record a phonetically-balanced script, optimized for coverage of the different sounds in English,” he added. In other words – they would bring someone in to record 30 hours or more of material, to make sure they get all the right sounds. With me, they had to improvise, but Leary said it worked out okay. “The radio reports are high quality and a suitable alternative – they are studio-recorded, read in a measured, consistent style and don’t have any interjections from other speakers, crowd noise, applause etc. that can make audio difficult to work with.” Trust me, this is a complicated process.
We've been working on a new voice for Washington DC correspondent @jamiedupree Jamie has blogged about it here: https://t.co/vreUMxxEce and there are some great reactions here: https://t.co/v3rEtoUUiN
— CereProc Ltd (@cereproc) June 12, 2018
4. Pairing the voice with a text-to-speech program. The folks at CereProc recommended downloading a freeware program called “Balabolka” to use with my Jamie Dupree 2.0 voice. While the name might be a tongue twister, the program is fairly straightforward. You load a specific voice to be used – in my case, the “CereVoice Jamiedupree – English (East Coast America)” voice. You type in some words. Then you hit the ‘play’ button. And it plays what you write. Hit another button, and it exports those written words into a computer generated audio file, either wav or mp3. Balabolka is a very powerful tool, and can probably do a lot more than I am using it for – but to see how it easy it was to hit Alt-W and generate an mp3 file with my new voice, it was really quite a surprise. So, when you hear me on the radio with this synthesized voice, it will just be me typing the words, and saving them into an audio file.
5. How does the voice work? When I type words into the text-to-speech program, it doesn’t go looking in an audio vault on my laptop for the exact words that I write, and then put those words together one-by-one. Instead, it searches out the sounds that would be made. So, this is not a question of having me on tape saying every word in the dictionary. Yes, it helps to have examples of me saying “President Trump” or “Congress.” But I know there was no example in my stories of me saying “Rudy Giuliani,” and yet, that popped out perfectly when I tried out the voice. How can that happen? CereProc uses “neural networks” to generate voices. “The neural networks, which contain between six to 10 layers each, work by slicing audio recordings of words down to phonetics,” the BBC wrote in a technical story about my new computer generated voice. This allows the Jamie Dupree 2.0 voice – and other voices created by companies all over the world – to navigate through just about any piece of text.
On Monday, veteran radio journalist @jamiedupree will return to the air with a new voice generated by artificial intelligence, after 2 years of silence 👨💻 Fascinating to see #NeuralNetworks applied to sound by @cereproc.
Here's my piece for @BBCTech: https://t.co/hXpOUf4Ga7
— Mary-Ann Russon (@concertina226) June 15, 2018
6. Figuring out certain words and sounds. While I have great praise for CereProc, the Jamie Dupree 2.0 voice isn’t perfect. One thing you run into immediately is that certain words and phrases don’t sound right – either because they are not pronounced clearly enough, or they seem artificially shortened. So, I spend a lot of time going back and moving words around in my news copy to see if it will sound better. One other way to massage the voice is that there are also a series of XML commands which can be used to emphasize certain words, to change the pitch, or alter the speed. One thing I quickly noticed is that the voice cuts off a word rather sharply at the end of a sentence – I simply found a way to fix that by slowing down the speed of the last word (or syllable) by 1 or 2 notches, to make it sound more natural. But there are some words that just don’t come out right, even if they are spelled correctly, so you have to be inventive. “Investigation” just doesn’t come out right, no matter what I try. House Speaker Paul Ryan’s last name didn’t sound good at all – so I wrote “Rye Inn” instead – and that sounded just right. Let’s take the word “denuclearize.” It sounded awful when written that way – but I found a way to make it sound better, as shown in the graphic below, by making it D-nuclear-rise, and by slowing down the final syllable.
7. What does Jamie Dupree 2.0 sound like? Listen to this extended ‘interview’ that I did with the BBC World Service. Some of the words and phrases sound natural and fairly normal – at other times, it gets a bit robotic. But to me, it’s still pretty amazing. It is my voice in there. And to be on the BBC World Service was a treat – I got hooked on shortwave radio as a teenager, and loved listening to Alistair Cooke’s ‘Letter from America’ each week. Will this voice solution work in the long run? That will be up to my bosses – and really, up to the listeners. If they can deal with the different sound – whether in a newscast, or a longer form appearance – then I will still be able to deliver the news from Capitol Hill. I fully expect to get a lot of people saying nice things, and I fully expect to get a lot of mean and nasty social media messages as well.
Two years ago radio reporter @jamiedupree lost his voice from a rare disease. A company has used his old audio recordings to develop a computerised voice – allowing him to talk again. 📻 https://t.co/lxSoAyO79C pic.twitter.com/oDCnuH3sZ6
— BBC World Service (@bbcworldservice) June 15, 2018
8. Comparing the old, the new, and 2.0 After two years of not having a voice that was ready for a trip to the grocery store – much less going on the radio – it is truly fantastic to have a way to get back on the radio. Yes, the voice is a bit robotic at times. But it is me. I can hear myself in these words. So, let’s look at how I sounded before, what I sound like now, and what Jamie Dupree 2.0 is like.
This news report is from February 28, 2016, at a Trump rally in Alabama. It was a giant crowd, and was one of my favorite reports from the first three months of 2016, when I was chasing the candidates all over the country.
What do I sound like now? I can get out words that sound okay, but not in any type of rapid fire way. If I am going to speak, it has to be very slow, and with a pen in my mouth to keep my tongue occupied (that is the source of my problem, a tongue which is not behaving properly, as it pops out of my mouth when I speak).
As you can tell from that audio, it is a struggle to say just about anything. So, we go to Jamie Dupree 2.0. It can say anything that I want (though four letter words don’t come out very well, just in case you were wondering). But, all I really want is to find my real voice again. Version 1 was better. But Jamie Dupree 2.0 is here, and this is what it sounds like.
9. Thanks to Rep. Ileana Ros-Lehtinen (R-FL). I can’t give any rundown on my voice without thanking Congresswoman Ileana Ros-Lehtinen. When she was elected after the death of Rep. Claude Pepper (D-FL), one of my company’s radio stations was in Miami, so I got to know her right away when she arrived in Congress in 1989. While Cox Media Group sold our news-talk station in Miami, I still kept in touch with Ros-Lehtinen in the hallways of the Capitol. She would hear me on the radio and happily chirp, “The most connected man in Washington!” When I told her of my voice problems late in 2017, she gave me a hug and said she would help. Her speech on the floor of the House in December drew attention to my problems, and spurred interest from news organizations. That’s how CNN’s medical unit got interested, and that’s how I found my way to Dr. Jinnah at Emory. I can’t thank Ros-Lehtinen, Speaker Ryan, and others for their help. It made a difference for me.
On @HouseFloor: Spoke about my friend @jamiedupree, a journalist who was stricken with Tongue Protrusion Dystonia. He is an example, for every American who has been afflicted with disease, of perseverance in the face of adversity. pic.twitter.com/LN0dMgc5hM
— Ileana Ros-Lehtinen (@RosLehtinen) June 12, 2018
10. How do I feel about Jamie Dupree 2.0? Let’s be honest. I want to be able to speak normally. Even just somewhat normally. A friend texted me to ask, was I nervous about the new voice? I guess, a little. But if there is one thing that I take from the last two years, it’s that I never gave up. I kept working at my job. I kept searching for a medical answer. I’m still searching for that answer. The outlook was admittedly bleak at times, like in April 2017 when the doctor at the Cleveland Clinic told me that no one could even treat my neurological/voice disorder. Early on, I knew I couldn’t give up. I have kids who are only 9, 11, and 14. “I think everyone saw how passionate and how badly/deeply you wanted this,” my boss told me the other day. “He never let anyone see him sweat,” said my friend and colleague Dorey Scheimer.
A few weeks ago, I was asked to come down to our company’s headquarters in Atlanta – our CEO Alex Taylor wanted to see me. It turned out to be an event with several hundred people, where I was presented with the “Governor Cox Award,” named after our company’s founder, Taylor’s great-grandfather, James M. Cox. Taylor told the audience that because of my voice problems, I could have given up, I could have gone on disability, I could have quit my job. But I didn’t. His words meant a lot to me, and they have been echoed by many inside our company in recent weeks. I want to thank him, and many others for their support.
Finally, I want to thank all the listeners, viewers, readers, and fellow ham radio operators who have sent me expressions of support over the past two years. Your words of encouragement were a great source of strength.
I would also thank those of you who sent me nasty emails, and celebrated my voice troubles. I know you will be back to criticize my new voice.
But you know what? Those jabs make me work even harder to stay in the news arena.
And now, we go onward – with Jamie Dupree 2.0.