Perceptual units in speech recognition.

Abstract

]The size of the sound stimulus employed in the first stage of speech processing was investigated in an attempt to determine the perceptual unit of analysis in speech recognition. It is assumed that the perceptual unit is held in a preperceptual auditory image until its sound pattern is complete and recognition has occurred. Vowels and consonant-vowel syllables were employed as test items in a recognition-masking task. The results show that recognition performance improved up to 200-250 msec, after presentation of the speech sound. The results were interpreted as evidence that the preperceptual auditory storage and perceptual processing of a speech sound does not exceed 250 msec., implying that some transformation of the speech signal must occur about every J sec. Since the stimulus within this time period must function as a perceptual unit, perceptual units appear to be of roughly syllabic length. The primary purpose of this study was to determine the size of the sound stimulus employed in the first stage of speech recognition. A listener has recognized (identified) a stimulus when he has determined that one of a possible set of alternatives was presented. Recognition of a stimulus is possible only if the information in the stimulus is sufficient to distinguish that stimulus from other possible stimulus alternatives, Recognizing speech continuously implies that each small portion of the sound-wave pattern uniquely determines a stimulus alternative. However, small portions of the acoustic input are not unique; there is no one-to-one mapping of stimulus to percept. Since the sound pattern must contain enough information for a consistent stimulus-percept mapping, larger chunks of the acoustic input are necessary for