Retrieving distinctive features and segments from variable acoustic cues
Kenneth Noble Stevens, Samuel Jay Keyser · The Journal of the Acoustical Society of America · 2000
It is assumed that words are stored in memory as sequences of bundles of distinctive features. Accessing words from the acoustic signal requires that these features be identified and organized into segments. Variability is a major obstacle to this process. Accounting for this variability requires a clear understanding of how variability arises when an utterance is generated by a speaker. It is proposed that variability arises from the operation of two principles: (1) Conservation: some of the defining gestures for a feature in a given segment may overlap with gestures from adjacent segments, causing weakening or masking of certain acoustic cues. (2) Enhancement: depending on the segmental and prosodic environment in which a feature occurs, additional gestures not directly specified by the feature may be introduced to enhance its perceptual salience. Conservation and enhancement provide a basis on which to select acoustic cues that reflect the gestures that produced them. Selection of acoustic cues for the features of consonant place, voicing, and nasality will be described and it will be shown that combinations of these cues can lead to feature identification with low error rate. [Work supported in part by grants from NIDCD.]