Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Combining Auditory Representations Using Fuzzy Sets

L. R. Gransden, Steve W. Beet

University of Sheffield, Department of Electronic and Electrical Engineering, Sheffield, UK

This paper addresses the problem of combining auditory representations of an acoustic signal based on a measure of the correlation of spectral events between representations in a way which attempts to preserve the detailed spatio-temporal events present in the auditory data. The fuzzy set theory class membership function is used to provide a measure of closeness of an observation vector to a set of centres positioned in the data space to characterise perceptually important spectral events. The technique provides a level of tolerance to the position of spectral events and allows a simple form of noise adaptation. A recognition experiment is described which demonstrates the improvement in noise robustness achieved for an isolated digit database at varying levels of signal to noise ratio (SNR) over a traditional MFCC pre-processor.-

Full Paper

Bibliographic reference.  Gransden, L. R. / Beet, Steve W. (1994): "Combining auditory representations using fuzzy sets", In ICSLP-1994, 1047-1050.