Size Perception for Acoustically Scaled Sounds of Naturally Pronounced and Whispered Words

Abstract

Recent studies have demonstrated that listeners can extract information about the length of a speaker's vocal tract from voiced vowels (Smith et al., 2005) and syllables (Ives et al. 2005). Smith and Patterson (2005) have also reported that they can extract size information from unvoiced vowels. In this paper, we extend the observations to words recorded in natural conversation mode to demonstrate that size perception is robust to changes in the mode of vocal excitation. We used a high-quality vocoder, STRAIGHT to produce scaled versions of voiced words with natural F0-contours, and scaled versions of unvoiced words that sounded like they were whispered. Size discrimination performance was measured for five reference speakers, using a two-alternative, forced-choice paradigm (2AFC) with the method of constant stimuli. The listener was asked to choose the interval with the word, or words, spoken by the smaller person. The just-noticeable-difference (JND) discrimination of speaker size was found to be about 5 % independent to the mode of vocal excitation. This value is a little greater than that reported in the previous syllable experiments and a little smaller than that reported in the vowel experiments. Moreover, the natural F0-contour in the voiced words and noise excitation in the whispered words did not affect the JND value. The results support the hypothesis that the auditory system segregates size and shape information at an early stage in the processing (Irino and Patterson, 2002).