An analysis of invariance in English stop consonants.
Kevin H. Richardson · The Journal of the Acoustical Society of America · 1992
Speech researchers have been attempting to isolate acoustic invariance in the speech signal for the past several decades. Currently, however, no conception of invariance has proven adequate to the task of demonstrating the existence of invariant acoustic features in the six English stop consonants. In the present study, static and dynamic versions of two recent invariance proposals [Sawusch and Dutton (in press) and Forrest, Weismer, Milenkovic, and Dougall (1988)] were developed and tested on a corpus of 1392 naturally produced stop and stop consonant cluster tokens. Classification accuracies of the metrics using stepwise discriminant analyses showed performance to be significantly different from chance but lower than expected had the metrics isolated invariance in the signal. Results will be discussed in terms of general human auditory-phonetic coding ability. Also, the notion that invariant acoustic features are the most useful way of conceptualizing the problem of phoneme recognition will be questioned. [Work supported by NIDCD Grant No. R01-DC00219 to SUNY at Buffalo.]