NMSU branding

New Mexico State University

New Mexico State University

News Center




Software created by NMSU engineering professor could improve sound enhancement

Suppose you were an undercover police officer trying to eavesdrop on a conversation between two drug dealers in a noisy club. Or suppose you were a 55-year-old executive with a hearing problem trying to understand what your spouse was telling you across a crowded room.



Dr. Phillip De Leon, NMSU Klipsch School of Electrical and Computer Engineering assistant professor, demonstrates software he has designed for sound separation. When fully developed, the software will have applications in the commercial, police and ...

Computer software being developed by New Mexico State University electrical engineering professor Phillip De Leon could help you in either situation by picking up an individual sound pattern and filtering out competing sounds.

Called "blind speech enhancement" for its ability to separate sound patterns without any preliminary information, the software could someday be part of audio equipment used by police, the military and consumers. It also could be incorporated into the microchip of a revolutionary type of hearing aid, De Leon said.

De Leon began researching the new program in 1999 with former NMSU electrical engineering professor James LeBlanc and continues the work with electrical engineering professor Krist Peterson. The Air Force Research Laboratories, whose Human Effectiveness Directorate is developing methods to improve human- to-human and human-to-machine communications, provided a two- year, $125,000 grant from 1999 to 2000. It awarded a second, follow-on, grant for $125,000 on Sept. 30, De Leon said.

What makes De Leon's software exciting to the Air Force -- and, potentially, to other users -- is its promise of immediate response and adaptability, he said.

Law enforcement and military agencies currently have methods of filtering out sound to make conversations clearer, but most involve obtaining a tape recording, then processing the tape later. In addition, while the technology is available to intercept on-going conversations, it typically requires positioning several expensive directional microphones so they pick up sound in a tight, narrow beam, De Leon said.

"This new method can work live and on the fly, and requires only a couple of relatively cheap microphones," he said. Also, because it is a computer program, as opposed to a hard-wired machine, the software can "learn," he added. "It adjusts, so the other voice -- or sound pattern, because the program can also be used with background noise or music -- eventually fades away without destroying the voice of interest. It only takes a couple of seconds to train," he said.

But, De Leon cautioned, the software is not yet fully developed. While currently capable of adapting to and isolating individual sound patterns, it requires a controlled environment.

The research required to make it work well in the outside world is what the second Air Force grant is intended to fund, he said.

One aspect of the new research is called the "under- determined problem," a situation where the computer is not given enough information to separate out the different signals, DeLeon said.

"Currently the method needs as many microphones as there are speakers, or sound patterns. In the new research, we clearly want to make the software perform well with just a couple of microphones, even though there may be many speakers," he said.

The toughest research problem is figuring out how to isolate a voice or other sound pattern in the presence of echoes or reverberation, he added.

"Right now the program can differentiate between different sound patterns, but the interesting thing about an echo is that it is really the same sound, but bounced off a wall, delayed a bit in time and made a little softer," he said.

Photo is available at http://kiernan.nmsu.edu/newsphoto/DeLeon.jpg.
For a print, call (505) 646-3221.
CUTLINE: Dr. Phillip De Leon, NMSU Klipsch School of Electrical and Computer Engineering assistant professor, demonstrates software he has designed for sound separation. When fully developed, the software will have applications in the commercial, police and military markets, he said. (NMSU photo by Michael Kiernan)

Jack King
Nov. 6, 2000