The role of STRAIGHT in research on the perception of size in speech and music

摘要

Fifteen years ago, while working on the mathematics of the gammachirp auditory filter, we realized that the perception of speech and music is largely scale invariant. People understand the speech of other people no matter what their average voice pitch or their mean formant frequency. People also know the family of an instrument (brass, string or woodwind) independent of its size and register. We illustrated how the auditory system could use a form of "stabilized wavelet-Mellin transform" to normalize the sounds of speech and music, and we decided to do some research on the perceptual invariance of speech and musical sounds. This was easier said than done, as it requires the manipulation of the acoustic scale variables in natural sounds. Fortunately, at about the same time, Kawahara-sensei released STRAIGHT which provided high-fidelity manipulation of the pitch and vocal tract length of speech sounds and musical tones. This paper describes a sequence of experiments on the perception of size using sounds in which the scale parameters were manipulated by STRAIGHT, and how the resynthesis element of STRAIGHT was adapted for musical sounds. The research provides one extended example of how STRAIGHT has empowered research on the perception of natural sounds.