Acoustic Invariance for Stop Consonants

Abstract

Previous research using synthetic speech has revealed that rapid frequency changes in the speech wave (transitions) are important cues for the perception of stop consonants. It has also been shown that these transitions are different for a given consonant in different vowel environments. Much emphasis has been placed on this finding, since the demonstration of acoustic invariance suggests that phonemes must be “abstracted” from the speech wave. The present research demonstrates that, for real speech, only a small initial segment of the transition is needed in order to correctly perceive all stop consonants. In fact, a small initial segment of each stop consonant contains all of the necessary acoustic cues needed to hear that consonant. For each stop consonant, we have been able to isolate a specific segment of sound, with most of the transition removed, which is correctly perceived in any vowel environment. For example, it is possible to record /di/ and /du/ on magnetic tape, remove the first 30 msec of each /d/, discard the second 30 msec, and splice the initial segment of either /d/ to either vowel. When this is done, subjects always report hearing a clear /di/ and /du/ These data show that there is an invariant set of acoustic cues for stop consonants.