By Mary Whitfill, News Staff
Looking to give a voice to the voiceless, Northeastern associate professor Rupal Patel and collaborator Tim Bunnell have launched The Human Voicebank Initiative, a crowd-sourced project aimed at creating personalized synthetic voices for those with speech impairments.
Using the voices of volunteers, Patel’s tool can merge certain parts of multiple voices into one unimpaired sound. While she has been creating synthetic voices for years, Patel is now attempting to streamline the process and make it easier for people to donate their voice.
The newest addition to the initiative is the debut of vocalid.org, where volunteers will be able donate their voices by simply recording two to three hours of themselves speaking into a relatively high quality recording device, such as a smart phone. Currently, participants must commit to a lengthy visit to Patel’s lab, but the goal is to make it as easy as possible for people to donate.
“VocaliD voices are made by blending the recipient’s residual voice characteristics with the speech of a non-impaired voice donor about the same age, size, etc.,” Patel said.
Patel partnered with Bunnell, head of the speech research lab at Nemours Biomedical Research, several years ago after his lab developed the ModelTalker text to speech system (TTS). However, ModelTalker is only a viable speech assistant to those who are losing the ability to speak, while Patel’s development can help those who have already lost, or never had, the ability to speak fluently. It is this discrepancy that made Bunnell so eager to get involved.
“Unquestionably, the most important goal is to provide non-vocal children and adults unique voices that do not sound like anyone else and do capture as much of their own residual vocal identity as possible,” Bunnell said. “I am especially interested in seeing this technology reach the kids [Nemours] serves, and of course, kids everywhere.”
The idea of synthesizing voices was born after Patel spent years researching different types of speech impairment.
“[I] documented that despite speech sound impairments, many people could still control the pitch, loudness and melody of their voice,” Patel said. “When I realized we could use their vocal characteristics to individuate synthetic voices – this project was born.”
Since giving a TED talk in Behrakis Health Science Center last December, Patel has received over 5,000 voice donor responses and has heard from hundreds of people interested in receiving a custom voice.
“People are just beginning to learn about this. At the moment, we are focusing on collecting donor voices so that we can make unique voices,” Patel said. “We are also collecting the names of people who want voices. Our plan is to spin this project out of the university so that we can address the scale of this effort.”
Roughly 7.5 million Americans have a voice disorder, according to the American Speech-Language-Hearing Association, and many of these people use the same computerized voice to communicate. The VocaliD project represents just one of many ways Patel hopes to revolutionize the lives of the vocally impaired.
“I recently received funding from Northeastern University to establish the Center for Speech Science and Technology, which will involve many students and faculty,” Patel said. “The VocaliD project will only be one amongst many other projects in the Center.”
The Human Voicebank Initiative has a goal of collecting one million voice samples, something Patel and Burnell hope to reach by developing a recording game or app to encourage people to donate their voices. Patel is an associate professor in both the Bouvé College of Health Sciences and the College of Computer and Information Sciences.