UMass Dartmouth Professor Karen Payton seeks to understand speech recognition

The voices on the answering machine and the ATM sound human, but aren't. The vocalizations made by many who are deaf don't sound human, but they are.

March 5, 2003 

UMass Dartmouth Professor Karen Payton seeks to understand speech recognition 

The voices on the answering machine and the ATM sound human, but aren't. The vocalizations made by many who are deaf don't sound human, but they are. Dr. Karen Payton, a biomedical and acoustics engineer at the University of Massachusetts Dartmouth, is fascinated by the challenge of making both simulated and impaired human speech intelligible, so she has spent the past ten years of her career probing the acoustical parameters of speech. 

She, herself, is amazed that it's been ten years since she embarked on this research during a research position she held at theMassachusetts Institute of Technology after she received her Ph.D. from The Johns Hopkins University. She is still a part of the MIT research team, spending her teaching and service hours at UMass Dartmouth, and conducting her research as a visiting scientist in the Research Laboratory of Electronics at MIT. 

"I don't have the answers yet, but I know that I'm not going to give this up," she says. 

The project that has so engrossed her is basic science, she explains, but when she and her colleagues have finished this work, its application will enable persons with impaired hearing to have better hearing aids and those ubiquitous voice machines to become more audible, more human-like. 

"I think hearing loss is the most devastating sense to lose, because all communication is based on it," she says. 

Payton's curiosity is piqued by the problem of isolating the variables that impact speech recognition. Did the listener understand what was said because they could intuitively predict the word sequence? Does the clear enunciation of radio broadcasters count more than speed for listeners? And how low can you go in a simulated bass register before you lose most people? 

Payton uses human subjects at MIT to calibrate what she terms "clear speech." To confound the linguistic wizards in her lab, Payton and her colleagues use nonsense sentences ("A right cane could guard an edge") so subjects can't guess at words they don't actually hear. And what she's found is that good diction is the most important factor in intelligibility—not tonal pitch or speed. What remains to be determined goes beyond what may appear to be the obvious: How do you measure "diction" quantitatively rather than qualitatively? Payton's team meticulously probes this question to tease out the solution. 

Although Payton does most of her work in this area at MIT, she is advising a UMD doctoral student and uses it in her UMD classes to teach undergraduate and graduate students applied acoustics methodology. And this past summer, she presented her research to a much wider audience at an international conference in the Netherlands. 

This article was written by Maeve Hickok for the News & Publications Office 


News and Public Information