HL14.OCT–AT&T–pm
DSP board recognizes voice commands
Up to 40 keywords can be stored for applications such as car phones and
microwave ovens
The Speaker-Trained Voice Recognizer points the way toward the next
generation of man-machine interfaces by allowing users of car phones to
dial a number by speaking the name of the person they wish to call. In
this particular instance, the voice directory is a matter of safety as
well as convenience. The board comprises 32-K x 8-bit, 150-ns static RAM
and a ROM-coded AT&T DSP16A processor with proprietary voice-capture and
modeling techniques. Interface logic allows a low-cost host microprocessor
to control the board. Tested in a car going at 30 mph with the air
conditioning on, the board achieved an accuracy rate of 95%. The keywords
can be up to 2 s in duration, and the response is less than 0.5 s. This
broadens the board's range of applications to include digital answering
machines, modem directory dialing, and industrial/domestic controls. The
board is supported by a PC-based development system with production
hardware and schematics and a base set of host control source software.
(DSP with ROM-coded software, $20 ea/10,000; PC-based development system,
$3,500–availability ???.) AT&T Microelectronics Allentown, PA Customer
service 800-372-2447, ext. 846; in Canada, 800-553-2448, ext. 846 EEM
FILE 2200
CAPTION:
The Speaker-Trained Voice Recognizer allows users of car telephones to
“dial” a number by speaking the name of the person they wish to call.
Advertisement
