Like many serial inventors, mathematician Dimitri Kanevsky looks for solutions for problems that he faces in his own life. In his case, some of his biggest challenges are related to the fact that he has been deaf since age 3.
Kanevsky, a member of the speech and language algorithms department at IBM Research, has invented a long string of hearing- and speech-related technologies. They include a system for helping people improve the effectiveness of lip-reading, a method that enables deaf people to converse on the telephone and an Internet-based system for capturing real-time transcripts of phone conferences. “I like to solve challenging problems, and I get a thrill from creating novel math concepts and making discoveries,” he says.
Today, Kanevsky will get another kind of thrill–when he’s honored with a Champion of Change award at the White House. The award recognizes individuals who make a positive impact on science, technology, engineering and math for people with disabilities. Here’s a livestream video link for the event.
While Kanevsky has a long record of achievements as an inventor, including 152 US patents, it’s clear from talking to him that some of his most important inventions may come in the future.
One area where his work could have a significant impact is in turning IBM’s Watson computer into a conversationalist. The brainy computer, which gained fame last year by beating two past-grand-champions on the TV quiz show Jeopardy!, translates text into speech, but, so far, doesn’t understand speech. Potential users of the technology say they would like to be able to converse with the machine in their work settings. Physicians, for instance, envision using IBM Watson as a well-informed adviser when they’re examining and talking to patients.
Kanevsky can help out. He and colleagues at IBM Research have contributed notable improvements to the Baum-Welch algorithm, which is used to make up for shortages of training data in speech-recognition systems. These days, they’re developing new methods for producing more accurate results in situations where people speak with heavy accents or use jargon, or where there’s a lot of background noise. These so-called discriminative algorithms could make possible real-time speech recognition systems that are fully automated and extremely accurate.
Watson is a first step in the emergence of a new era in information technology–which IBM calls the Era of Cognitive Systems. Big shifts are coming in chips, systems, data and programming. Kanevsky believes that these developments offer the promise of major breakthroughs in accessibility technologies–but only if those who architect them take people with disabilities into account from the start. “If this is done in the right way, it will have a tremendous impact on all people with disabilities,” he says.
One of Kanevsky’s earlier inventions, Artificial Passenger, is a system for keeping sleepy drivers awake by telling them stories and jokes, asking questions, and suggesting that the driver take a rest break if it detects that she or he is sleepy. Kanevsky came up with the idea after keeping his wife, Galina, awake by talking to her on a late night drive.
Watson’s a whole lot smarter than Artificial Passenger. For example, Watson can digest the universe of published information about a particular topic and then answer open-ended questions about it. So adding conversational capability to Watson would take speech recognition to a whole new level. Systems can be created that understand both speech and meaning. That’s a worthy challenge for an ambitious serial inventor.