Communication Scientist, Engineer Craft Voice Diagnosis Tool

November 6, 1998

GAINESVILLE — It’s a kind of CAT scan for professional singers’ voices.

A University of Florida communication science professor is using a UF College of Engineering-designed computer program to sort out the myriad qualities that make singers’ voices shine. The goal is to identify signs of trouble in time to help singers correct them.

“With a singing voice, I’m interested in trying to objectively determine what makes up an exceptional voice,” said Howard Rothman, professor of communication sciences in the College of Liberal Arts and Sciences. “If I knew that, I might be able to identify and help singers who are having difficulty early enough that their careers could be saved.”

Fans of opera or other vocal musical forms, such as liturgical Jewish music, may think they know what makes singers’ voices good, but impressions are misleading, Rothman said.

One of his studies showed singers, speech pathologists, voice coaches and other experts in speech and music could not collectively identify singers’ vocal techniques. For example, more than 100 experts sampled had different opinions on whether singers were using vibrato — slight and rapid variation in pitch — or other techniques such as tremolo.

There was a little more agreement on whether a singer sounded good or bad, but opinions remained far from unanimous, Rothman said.

“We’re really not sure what the dimensions are that define these terms,” he said.

Currently, when singers want to correct problems with their voices, they must rely on coaches or others whose prime tool is their subjective listening abilities. The computer program opens the possibility singers one day will have an additional, objective tool to gauge their problems.

“It’s opening the voice up to minute inspections that are not human biased,” said Tony Arroyo, an associate professor of electrical and computer engineering and director of UF’s Machine Intelligence Laboratory. Arroyo supervised a recent UF graduate, Jose Antonio Diaz Avila, who designed the program for his dissertation in electrical and computer engineering.

Singers’ voices are composed of tones with different frequencies, and the computer program separates the tones from one another and renders them as separate waveforms.

The program then allows Rothman or other researchers to pick the waveform they are most interested in and dissect it on a computer screen. By examining many samples, researchers can identify the mathematical characteristics of each singer’s voice, Arroyo said.

“For example, we could come up with a formula that describes Pavarotti’s vibrato,” he said.

Rothman said the program allows him to examine 149 parameters of the human voice, though he so far has confined his investigation to just six. It also has made the process much faster, he said. “In less than five minutes, I can get information that used to take me hours and hours and hours to develop,” he said.