Creating Artificial Voice With Neural Network
Jamie Dupree is an American radio journalist, who covers politics. Due to a rare neurological condition, he lost his voice two years ago. Now, at 54, Dupree is glad he can make a come back thanks to artificial voice created by CereProc’s neural network.
Dupree was an active political journalist, who produced news stories for six radio stations. He was an important source of news from the Congress in Washington DC until 2016. Two years back, his voice began to fade due to a rare neurological condition called tongue protrusion dystonia, which affects both the tongue and throat of the patient making it difficult for them to speak more than a few words at a time.
Free Step-by-step Guide To Become A Data Scientist
Subscribe and get this detailed guide absolutely FREE
Dupree did not give up his career, however. He interviewed policymakers by writing questions on an e-writer tablet and recording answers in one-to-one or group interviews. But, he had to go off-air from the radio. His stories were published instead of being narrated.
The situation caught the eye of an US Congress member, who spoke about Dupree in the House of Representatives. The speech drew attention to Dupree’s condition and his employer decided to actively seek for a solution to return Dupree his voice back. They had files from around 30 years of the radio journalist’s broadcasts, which was a useful resource in this endeavor.
Creating Artificial Voice
Dupree’s artificial voice is now created by a Scottish tech company called CereProc using a neural network. The computing systems created by the company use samples of Dupree’s voice from old recordings to replicate and recreate his way of taking. The artificial voice has saved Dupree’s career. One, obviously, cannot be a radio journalist without talking. It has given him his job back and protected his family from financial troubles.
Generally, it takes data from about 30 hours of recording to create an artificial voice of someone. It is done by chopping up words in audio files and putting them back together as per requirements using artificial intelligence. Advancements in artificial intelligence allows for predicting and imitating of individual speech patterns if sufficient data is available.
Creating an artificial voice was expensive and time-consuming in the past. It took tens of thousands of pounds and about a month’s time to create just one artificial voice. With CereProc’s neural network, however, the process has become more affordable and less time-consuming. Cereproc can create an artificial voice for as little as £499 in just a matter of days.
How it Works
To create your own artificial voice, all you need to do is read a script on CereProc’s website. The company’s neural network then slices every word that you read into a hundred or so pieces until sufficient knowledge is gained of your particular use of phonetics. Using this knowledge, the neural network can predict what you would sound like in conversation to create your voice.
The use of neural networks for image recognition is already quite popular. CereProc believes, however, that applying artificial intelligence and neural networks to sound is much easier. According to CereProc’s Co-Founder & CTO Chris Pidcock, “AI techniques work quite well on small constrained problems and learning to model speech is something deep neural nets can do really well”.
Gift of AI
Dupree is expected to make a return to radio next week on 25 June. His computer-generated artificial voice will be heard by listeners of WSB Atlanta and stations owned by Cox-Media in Tulsa, Jacksonville, Orlando and Dayton regions of America.
With the artificial voice created by CereProc, Dupree can simply use a program named Balabolka to turn his script into narration. If something doesn’t sound right, he can adjust the settings to change consonants and vowels or even words! In just about seven minutes, he can create a complete radio story and go live with it.
Listening to his artificial voice, Dupree said, “It is me, there is no doubt about that.” He thinks it is a little robotic, but he wasn’t expecting it to be perfect anyway. Dupree still mostly relies on e-writer to communicate with family, friends and colleagues.
Dupree likes to write blogs, stay active on twitter and use Facebook. But nothing gives him more joy than his new-found voice, which he can use to narrate a 20-second story in the beginning of a newscast. Safe to say, artificial intelligence has truly made a significant difference to this radio journalist’s life.