The University Record, April 30, 1996
Speech pathologists may get new weapon to fight stuttering
By Randy Frank
U-M-Dearborn
The signs are clear—involuntary repetition of words or syllables, or hesitation before saying a word. The problem is stuttering, and it occurs in 1 percent of the U.S. population. If the problem isn’t treated in early childhood and persists into adolescence, stuttering is likely to be life-long.
Speech pathologists, though, may soon have another weapon to fight stuttering, thanks to an effective and inexpensive speech therapy tool developed by Louis Przebienda, a U-M-Dearborn graduate student in electrical and computer engineering.
For his master’s thesis, Przebienda, a hardware design engineer at Acromag Inc., wrote a computer program to help people who stutter.
Przebienda’s research originated from a group project in his graduate class, Electrical and Computer Engineering 584: “Speech Process,” which sought applications of speech processing and digital signal processing technology.
The research project grew from a collaboration forged by Selim Awad, associate professor of electrical and computer engineering, and Richard Merson, a speech pathologist at Beaumont Hospital in Royal Oak, who sought to create speech therapy tools using digital signal processing technology.
A demonstration of speech therapy tools by Merson solidified Przebienda’s interest, inspiring him to create “a speech -fluency shaping aid.” His thesis project was supervised by Awad.
In Przebienda’s program, speech therapy clients can view a graphical display of their irregular speech patterns and contrast it with their clinician’s speech patterns through computer-generated audio and visual cues.
“The children (clients) love it. They can see the shape and amplitude of their utterances,” according to Merson, who chose the sample utterances for the program, which is used primarily by children ages 7 to 13.
First, a speech pathologist must record a number of words or short phrases, up to six seconds long (about 20 to 25 syllables), via the PC’s microphone and sound card. Short utterances, vowels or consonants, as short as 30 milliseconds, also can be recorded.
When clients are at the computer, they select a pre-recorded word or phrase, Przebienda said, then listen and view it on the computer screen while they repeat the phrase into the microphone.
After speaking a given phrase into the microphone, the client’s “average magnitude profile,” the primary source to measure fluency and performance, is calculated and displayed graphically by superimposing it onto the clinician’s magnitude profile of the same phrase. The client and clinician then can compare the two graphs.
The results of the magnitude calculations are analyzed for start and end alignment, onset, amplitude and shape matching. After the analysis, a score is assigned to the client’s performance in each of these categories, Przebienda said.
“A low score means the client had episodes of stuttering,” he said. “The software allows the client to practice ‘model’ speech samples that range from vowels and syllables to complete sentences at various levels of volume, rate and sound duration with computer visual feedback. They learn to alter their speech sound, word rate, loudness, duration and onset to initiate and maintain speech fluency.”
One advantage of Przebienda’s software is that it duplicates the function of complicated and expensive acoustic equipment available only in well-equipped speech language pathology clinics.
Since the software allows individuals who stutter to continue their treatment at home, the cost of therapy is reduced and the speech pathologist is available to see other clients, according to Przebienda.
In the future, Przebienda plans to improve the scoring and feedback features, and make the program more user-friendly. He also will integrate the software with hardware developed by Brett Bentzinger, a U-M-Dearborn electrical and computer engineering undergraduate, that will enable the program to run in real time, visually displaying speech patterns as clients speak into the microphone.
