Perception of the voiced-voiceless contrast in syllable-final stops

A computer editing technique was used to remove varying amounts of voicing from the syllable-final closure intervals of naturally produced tokens of /p epsilon b, p epsilon d, p epsilon g, pag, pig, pug/. Vowels for all six syllables were approximately the same duration, and the final release bursts were retained. Identification results showed that voiceless responses tended to occur in relatively large numbers when all of the closure voicing and, in most cases, a portion of the preceding vowel-to-consonant (VC) transition had been removed. A second experiment demonstrated that removal of final release bursts had very little effect on the identification functions. Acoustic measurements were made in an attempt to gain information about the acoustic bases of the listeners' voiced-voiceless judgments. In general, stimuli that subjects tended to identify as voiceless showed higher first-formant offset frequencies and shorter intensity decay times than stimuli that subjects tended to identify as voiced. However, for stops following /i/ and /u/ these acoustic differences were relatively small. We were unable to find a single acoustic measure, or any combination of measures, that clearly explained the listeners' voiced-voiceless decisions.