|
You are at: Wagner
Home > Technology > Voice-Speech > Speaker-Verify
Speaker Identification and
Speech Recognition Solutions
Army
SBIR Phase II Quality Award
Our Phase II Small Business Innovation Research (SBIR) project,
"Continuous Speech Recognition with Speaker Verification
for Secure, Real-Time Voice Control" has been selected as
one of the five winners of the 1997 Army SBIR Phase II Quality
Award.
Selection was based on the following criteria:
- Originality and innovation
- Relevance to the Army, and
- Immediate commercialization potential
ARPA (precursor of DARPA)
Phase I SBIR
Under a Phase I
SBIR effort for ARPA completed in May, 1994,
Daniel H. Wagner Associates developed an automated speaker authentication system designed
to continuously verify a speaker's identity. Two systems were
developed in this effort and each performed at levels close to
that demonstrated by humans for the same task on the same test
data.
The results produced by these systems are comparable to the
best speaker identification/verification systems available today.
Signal processing schemes used in the work included: linear predictive
coding, affine wavelet transform methods, and generalized time-frequency
representation techniques.
U.S. Army Phase I SBIR
Under a Phase I SBIR effort for the U.S. Army completed in
October, 1994, Wagner Associates developed an automated speaker identification
system. This system was designed to automatically identify a
speaker from a given group of speakers using only the sound of
his or her voice (text independent). The technology developed
here has applications in both security and command and control
applications.
Our work was focused on increasing the reliability of verifying
a speakers identity while minimizing the amount of time needed
to perform the verification. Building on our previous ARPA effort,
we were able to perform significantly better than 90% verification
accuracy using only approximately 0.1 seconds of speech data.
Further, only a few seconds of speech are required to
determine and store speaker characteristics.
U.S. Army Phase II SBIR
Under a Phase II SBIR effort for the U.S. Army beginning in
March, 1995, Wagner Associates expand on our previous work by addressing the
combined problem of speech recognition and speaker verification.
Here, we are developing a prototype machine control system in
which several different robot systems will operate under voice
control from several speakers. The control system must recognize
what has been said, determine if the person who gave the command
is an authorized speaker, understand the command, and
then relay the command to the appropriate robot system.
Specific applications of such technology include voice control
of an autonomous vehicle or of a tank subsystem. There, the control
system would listen to a conversation among several speakers
(e.g., the tank crew) and obey only those orders given by personnel
authorized to command that vehicle/subsystem.
Our system maximizes the use of Commercial Off-The-Shelf (COTS)
hardware. Wagner-developed software is integrated with these
components. The prototype system will perform the following tasks
simultaneously and in near-real-time for multiple speakers:
- Automatically recognize continuously spoken commands from
a large vocabulary in noisy environments using preregistered
speakers
- Automatically verify the identity of the speaker without
repetition
- Provide feedback on spoken utterances, speaker identity,
and system status
- Transmit commands compatible with Army robotics and decision-aid
systems over an appropriate computer interface.
|
