New technology blooming

Richard Rose

For all those who feel frustration at pushing tiny buttons on increasingly diminutive devices: Richard Rose is intent on making your voice heard. "We're getting internet access from increasingly tiny cellphones with impoverished displays and limited input modalities," says Rose. "Voice is the best way to interact with these devices; that, to me, is a compelling application of speech recognition." And automatic speech recognition, the tricky business of getting computers to understand human utterances, is the new Electrical and Computer Engineering professor's realm of expertise. "Getting a mobile device to recognize speech involves considering a lot of additional issues, such as the fact that when I'm on the move with a mobile device it will be noisier, and there is a wider variety of devices a mobile system will have to interact with," he explains.

Owen Egan

While automatic speech recognition has experienced its success stories, we may still have to wait some time before we can bark orders at our BlackBerrys. The vagaries of the human voice remain a challenge. "It's hard to characterize the state-of-the-art in a general way," Rose says. "I can think of four companies in Montreal that design automatic speech systems, and they work well because the systems are for particular domains — helping you find your luggage, helping you fix problems with Bell Canada, helping you make airline or train reservations." But even with a clearly defined domain, there are limitations. For instance, Rose notes, "A system based on European French will fall apart on Quebec French." Automatic speech recognition these days ranges from that used in telephone systems to incredibly complex dictation-transcription programs, though the latter remain especially challenging. As well, now that closed captions are legally required for a certain percentage of television broadcasts, there is a concerted research effort to render the process automatic.

Rose's doctoral research at the Georgia Institute of Technology focused on the compression of speech for transmission over narrow-bandwidth channels, and after graduation he found himself at MIT. "They had a very strong group in automatic speech recognition, and I just drifted in," he explains. "I was interested in modelling techniques and their applications, and there was a good research community. It was a 'right time, right place' scenario." From MIT he migrated to industry, establishing himself at AT&T Labs before coming to McGill on January 1, 2004. There are plenty of groups here for Rose to participate in, such as the Centre for Intelligent Machines and the new Centre for Advanced Systems and Technologies in Communications (SYTACom). He is also forging links with the local industry.

The biggest change in the shift from industry to academia involves that unique element that distinguishes the two: students. "I'm just now adjusting to the teaching, and it's quite a transformation," says Rose, who took on the odd graduate student while at AT&T Labs. "One of the things that attracted me to the university life was the process where you start with students who don't know anything at all, and you wonder if the time you spend with them will ever pay off, and then you get to the point where they are the experts," he says.

And the easiest transition? "Montreal — what a great city," he enthuses. "I lived in New Jersey when I worked at AT&T Labs, and we would hang out in New York. Montreal is very diverse, and all that diversity seems to be tied together a little bit better than in New York." Even with — or perhaps because of — the wide range of accents.

Patrick McDonagh