"Hearing aids of the future will be required to not only amplify sounds in different frequencies differently, but they will also need to help people gain auditory focus so that they can sit in an environment and feel that they have the ability to look at somebody and listen to this or that person with ease."
Associate professor of electrical engineering
"Thank you for calling. Say '1' for this department, say '2' for that." You lean into the phone and try to speak clearly, hoping the machine will understand.
This is an example of voice-to-text technology, an application of speech signal processing that Professor Robert Nickel, electrical engineering, describes as "getting machines involved in transmitting, storing, recognizing and synthesizing human speech." He says the automatic recognition of a few specific words can be done fairly well. However, the process becomes very difficult and subject to large errors when an unrestricted vocabulary is involved.
Researchers began to work on automatic speech recognition about 60 years ago with the expectation of full voice-to-text capabilities within a couple of years. The problem turned out to be much more complex than they thought. Nickel explains, "The problem is insanely difficult. The more you get into it the more you realize that understanding speech and what's behind it is really something that you cannot just work on at the acoustic level alone, because what makes people understand speech are the cognitive processes in the brain that come to our aid to patch up the engineering catastrophe we call speech." Nickel is trying to bridge this gap between acoustics and cognitive processes by working on noise reduction on the front end.
"Whenever you record somebody's speech there are all kinds of ways in which the signal is not always the same," he says. "On some days I speak one way and on other days I speak slightly differently. In addition, there are different environments. Then there is background noise. It is very easy for a human being to separate these things out, but it is very difficult to do that for a computer."
Currently, Nickel is working on hearing aid functionality. It can be exhausting for someone wearing a hearing aid to separate out all the sounds they hear from the sounds they really want to focus on. "Hearing aids of the future will be required to not only amplify sounds in different frequencies differently, but they will also need to help people gain auditory focus so that they can sit in an environment and feel that they have the ability to look at somebody and listen to this or that person with ease," he says. "I'm working to develop algorithms that will allow us to enhance the voice of a targeted speaker while, at the same time, suppressing background noise and dampening other voices.
Nickel's latest work was published this May, but the projects could be years away from implementation: Since the algorithms run on a mainframe computer, the hearing aid would have to be attached to a car battery sized computer to work. "This highlights the difficulty and challenge of the problem," he says.
Posted October 3, 2013