|
|
|
| Current Issue |
|
|
| Resources |
|
|
|
Machines That Hear
| Article
# : |
11795 |
|
|
Section : |
NATURAL SCIENCE
|
| Issue
Date : |
4 / 1987 |
2,672 Words |
| Author
: |
Phil Cohan
Phil Cohan is a free-lance writer who lives in Washington,
D.C. |
The house looks typical, but seems to be alive as its lone female occupant is apparently talking to a wall. By uttering a few words, she turns up the heat in her home; has a bath drawn to a specified temperature; obtains answers from the computer about the weather outside; turns on a coffee-brewing machine; and activates a security system.
The ordinary-looking house is now under construction in Bowie, Maryland, by the U.S. National Association of Home Builders (NAHB). Utilizing the latest technologies, this experimental house will contain a series of voice-activated control devices capable of a wide variety of responses.
The NAHB plans to build fifty to one hundred of these "smart" houses by the end of 1989 for demonstration and testing. Then, in 1990, they will be marketed in the U.S. Eventually, the price of the equipment is likely to decline so that even moderately priced homes will include some "smart" features. The Yankee Group, a research firm in Boston, forecasts that almost three million homes in the United States will contain some of these features by 1996.
At AT&T's Bell Laboratories scientists have built a speech-controlled robot that responds to a fifty-one-word vocabulary of commands. This speech-recognition system allows "connected speech," requiring no awkward pauses between words. The fifty-one words enable use of 1.2 million different valid command sentences up to fifteen words in length. The robot uses an ultrasonic rangefinder to find objects on a work table, and then queries the operator as to the description of the size, shape, and color of the object. When the operator responds in normal conversational speech, the robot learns the description of each object so that it can later pick up that object on voice command.
IBM researchers are working with an experimental desktop computer with a 20,000-word vocabulary that takes dictation. In the realm of speech recognition by machines, this is a major step forward, but the system is still far from attaining the dream of scientists in the field. They envision a machine that can understand the language as well as a human, including a large vocabulary of connected speech spoken by a variety of people. In comparison to that vision, the IBM machine is limited in the following ways: 1) the words cannot be spoken continuously; there must be a break between them, and 2) the computer must, in advance, be calibrated to each person's unique speech pattern.
Multidisciplinary
...
Read Full Article
Look for this article in Ask.com
|
|
