Perception of English consonants from dynamic acoustic properties

Abstract

Phonetic perception is often said to depend on short-term attributes of vocally produced sound. Counterevidence to this premise is found in the example of sinewave replicas of speech. The dynamic properties of a sinewave replica evoke phonetic impressions despite the absence of typical or natural sounding short-term elements. To gauge the robustness of phonetic perception from dynamic acoustic properties, we compared identification performance by normal listeners across identification tests composed of signals that preserve dynamic properties of speech despite the absence of short-term speechlike properties: sinewave replicas modeled on the natural samples of 18 English consonants in CV syllables, noise-band vocoded signals derived from the natural samples, and chimerical signals exhibiting the coarse-grain spectrum envelope of natural speech and the fine structure of nonspeech signals. Identification from dynamic acoustic properties was well predicted from control performance using sampled natural speech in noise, showing that short-term acoustic properties and their corresponding auditory qualities are inessential in perceiving consonants. This outcome is consistent with a perceptual account that favors attention to the characteristic modulations of an acoustic carrier, and opposes an account appealing to auditory essences of each phonetic element.