http://www.delawareonline.com/apps/pbcs ... /-1/NEWS01
By IN-SUNG YOO
The News Journal
Johanna Kelly is the essence of bubbly youth, all smiles and full of energy.
But when the 6-year-old Brandywine Hundred resident introduces herself through her portable speech generation device, the halting voice emanating from the box doesn't seem to fit.
"My name is Johanna. I go to school at Lancashire," the device says as Johanna selects pre-programmed phrases.
The voice is a young girl's, but it's flat and has a slight echo, the intonation unnatural.
Johanna, who has an unidentified developmental delay, cannot speak. Her only "voice" is that of an augmentative and alternative communication device, or AAC. Like most AAC users, Johanna doesn't have much choice when it comes to her simulated voice. But a project under way at Alfred I. duPont Hospital for Children in Rockland may one day give her a voice that more closely matches her appearance and personality.
The ModelTalker System -- developed by Timothy Bunnell at the Nemours Center for Pediatric Auditory and Speech Sciences in cooperation with the University of Delaware -- is a speech synthesis program for people with health conditions that affect their ability to talk.
Unlike the software used in most AACs, speech programs like Model- Talker take real recorded speech, chop it up into its individual sound components, catalog the sounds in a database, then piece them back together as needed. The program further tweaks the results to improve intonation. About 1,600 sentences and isolated words are needed to make up each voice profile. Users "speak" by entering text into a screen through a keyboard or other pointing device.
Currently, almost all AACs in the world use one of a handful of 1980s-era speech synthesis systems, Bunnell said. The DECtalk program -- perhaps best-known as the artificial voice of renowned British physicist Stephen Hawking, who is almost completely paralyzed by Lou Gehrig's disease -- is the most popular software of its kind and quite effective. But for many users, the "robotic" quality of the voice can be off-putting. And the lack of variety in voices allows for little in the way of individuality.
"Everyone has a very strong vocal identity, except for users of AACs, who all sound like DECtalk Paul," Bunnell said. "You get a room of people all using AACs together and it can get pretty confusing and sort of annoying, both for the users and everybody standing around, because everybody speaks with the same voice."
The main goals of the ModelTalker -- the only speech synthesis program of its kind being developed for AACs -- are voice banking and voice donation. For people who are likely to lose their voice from the progression of disease, Model- Talker offers the ability to store their vocal characteristics in digital form. When they can no longer speak, their own voice has been preserved.
For people who were born with conditions that kept them from ever learning to speak, Bunnell hopes ModelTalker will one day allow them to use a donated voice from a family member or friend, or from a catalog of prerecorded voices. Nemours is working with Newark-based AgoraNet to develop a finished product that may be ready as early as mid-2008.
There's definitely interest in technologies like ModelTalker among the speech-impaired community, said Jennifer Brand, director of patient services at the ALS Association, which helps in the fight against amyotrophic lateral sclerosis, or Lou Gehrig's disease. Improved voice quality helps users of the technology get through their everyday tasks more efficiently, but it also means a lot for them to be able to communicate with friends and family without the distraction of an unnatural voice.
"It does make a big difference to the patients," Brand said. "It makes them sound more human."
For now, Johanna relies mainly on sign language to communicate with her parents and teachers. At school, though, her AAC allows her to talk to teachers and even participate in sing-alongs by playing pre-programmed songs. But Cathy Kelly hopes the ModelTalker will one day give her daughter a unique, more realistic voice, one less likely to draw derision from some of the meaner children.
"She has two sisters, so if we could record one of their voices for her, something like that would be nice," Kelly said.