A watch that understands English? 'Artificial listeners' surge ahead
What do you say to a watch? Most clear-thinking folks don't say anything. But if the Swiss watchmaking group Asuag AG has its way, that may soon change. In its labs, the company is developing a timepiece fitted with a microchip that will respond to 15 word commands in several different languages. To set the alarm, for instance, you might say, ''Alarm go off at 6:30.''Skip to next paragraph
Subscribe Today to the Monitor
This development will probably not bring a giant leap in watch sales, but it does represent one more small tick in the longstanding drive to devise machines that better understand human speech.
Voice-recognition machines have long been used in industry for quality control, inspection, and simple data entry. Instead of two people being required to spot flaws in, say, a circuit board - one to examine the part and the other to tap the information into a keyboard - one inspector can do it by simply talking into a microphone. The information is automatically logged into a computer fitted with a special voice-recognizing electronic circuit.
Now a wave of speech-recognition consumer products are near that will make it possible to give voice commands to personal computers, video games, and, perhaps , telephones.
''We are at a real turning point in the state of the art,'' says Janet Baker, president of Dragon Systems Inc., a Massachusetts speech-recognition research firm. ''There is finally getting to be a good match between technology and performance.''
The ultimate aim is to develop ''artificial listeners'' keen enough so that anyone can talk to the machines in conversational language and be understood. This is a goal some believe will never be achieved, though scientists are edging toward it.
The problem lies in the complexity of human speech. For one thing, such a machine would have to have an almost unlimited vocabulary. For another, it would have to tell when each word in a sentence begins and ends - easy for man but staggeringly hard for machine. There are also the variations in voices - in tone , pitch, and accent - that the machine has to sort through. Background ''noise'' is a problem, too.
Present systems, as a result, are limited in what they can do. Typical circuits, often costing several thousand dollars, understand only a small vocabulary (usually less than 350 words) - and then from only one or two speakers. Systems exist that can decipher many different voices, but only when a handful of words are used.
These drawbacks, coupled with a limp economy, have created a less than robust market for artificial-listening devices, prompting several companies to pull out of the fledgling field in recent years.
Still, industry analysts expect the voice-recognition market - now around $20 million a year - to hit $1 billion by 1990. A variety of new products are about to give the industry a boost. Texas Instruments is soon to come out with an estimated $2,500 voice attachment for its personal computer to process ''continuous'' speech commands. Verbex, a division of Exxon Corporation, recently unveiled a circuit that recognizes several hundred words of ''connected'' speech, spoken without pauses. By year's end, the first voice-control video games are due on the market.
Further out, automakers want to design voice-control circuits for everything from air conditioning to wipers - presumably to give drivers more time to watch the road. Prototype voice-dialing telephones have been developed. So have appliances that flip on at the touch of a word. Analysts, however, expect things like voice-control TVs and ''listening'' watches to appeal mainly to the gimmick-minded - although there may be a significant market among the handicapped.
Promising, but still far off, is the possibility of a ''voicewriter'' - the science-fiction typewriter that instantly turns words into perfect, neat business letters. IBM is working on one. But it is at least two years and several million dollars away in the lab, says IBM scientist Frederick Jelinek. A store-shelf version, if developed at all, may be a decade off.
What all this means is that voice-recognition devices - as limited as they are now - will probably continue to make inroads on the factory floor and to a lesser extent in the home and office. But the days of ''talk-typewriters'' and humanlike voice machines are still a long way off. Here, as speech-recognition expert Roberto Bisiani of Carnegie-Mellon University points out, one thing is clear: When it comes to listening, man still has the edge over machine.