Data processing: artificial intelligence – Neural network – Learning task
Reexamination Certificate
2000-05-31
2003-05-27
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Neural network
Learning task
C706S015000
Reexamination Certificate
active
06571229
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to classification systems, e.g. speaker recognition systems, and more specifically to a method and apparatus for iterative training of a classification system.
BACKGROUND OF THE INVENTION
Modern classification systems require high accuracy training for optimal performance in a variety of environments. One method of achieving high accuracy is through discriminative training methods. A discriminative polynomial classifier for speaker verification is described in detail in W. M. Campbell and K. T. Assaleh, “Polynomial Classifier Techniques for Speaker Verification”, in
Proceedings of the International Conference on Acoustics, Speech, and Signal Processing,
pp. 321-324, 1999. Iterative techniques to solve linear equations have typically been used in two areas. In the numerical analysis community, methods are targeted toward solving large sparse systems. In the engineering community, approaches have concentrated on using iterative methods for recursive learning. The present disclosure applies to both areas.
Polynomial discriminative training methods optimize the performance of a classifier by maximally separating the decision regions. The main advantages of this polynomial approach are:
the training method is able to handle large amounts of enrollment data in a straightforward manner;
the architecture is based upon a simple multiply-add only architecture;
the classifier is trained discriminatively with an algorithm achieving the global minimum; and
the classifier output approximates a posteriori probabilities, which eliminates the need to perform cohort selection and cohort scoring (cohorts are incorporated as part of the training).
A major difficulty in using polynomial discriminative training for previous systems is the large memory footprint required for training. The training process requires the solution of a large (for small platforms) matrix problem. This is a serious drawback for portable devices, sometimes prohibiting discriminative training from being a viable choice. Many portable devices (e.g., cell phones) have high MIPS (i.e., they include DSPs and the like) but little memory. Therefore, it is desirable to construct methods and apparatus that minimize memory usage and produce equivalent functionality.
Accordingly the present disclosure describes a new and improved method and apparatus for iterative training of a classification system in which memory usage is substantially reduced while producing equivalent functionality.
REFERENCES:
patent: 5917953 (1999-06-01), Ausbeck, Jr.
patent: 0997828 (2000-05-01), None
Assaleh K T et al: “Speaker identification using a polynomial-based classifier” ISSPA '99. Proceedings of the Fifth International Symposium on Signal Processing and Its Applications (IEEE Cat. No. 99EX359), Proceedings of Fifth International Symposium on Signal Processing Symposium on Signal Processing and its Applications, Brisbane Qld., Australia, Aug., 22-25, pp. 115-118 vol. 1, XP002183435.
Fernandes P et al: “A New Storage Scheme for an Efficient Implementation of the Sparse Matrix-Vector Product” Parallel Computing, Elsevier Publishers, Amsterdam, NL, vol. 12, No. 3, Dec. 1, 1989, pp. 327-333, XP000081653.
Campbell W M et al: “Polynomial Classifier Techniques for Speaker Verification” 1999 IEEE International Conference on Acoustics, Speech, and Signal Processing, Phoenix. AZ, Mar. 15-19, 1999, IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), New York, NY: IEEE, US vol. 1, Mar. 15, 1999, pp. 321-321, XP000900123.
Davis George B.
Koch William E.
Motorola Inc.
LandOfFree
Method and apparatus for inerative training of a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for inerative training of a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for inerative training of a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3092273