Gross versus detailed spectral cues in spectrally distorted speech
Michael Kiefte · The Journal of the Acoustical Society of America · 2000
An experiment was designed to address the importance of gross versus detailed spectral features in the perception of stop-consonant place of articulation. This is done by presenting listeners with noise-vocoded stimuli in which the spectral resolution of the processed tokens is severely reduced. While it can be assumed that such a transformation will effectively eliminate detailed spectral features, such as burst-peak and formant frequencies, previous work has shown that this is not necessarily the case [M. Kiefte, J. Acoust. Soc. Am. 106, 2273 (1999)]. Nevertheless, it is shown that gross spectral features, such as spectral tilt and compactness, are better able to predict listeners’ identifications of noise-vocoded stimuli. While this suggests that listeners attend primarily to gross spectral shape cues, which are assumed to be particularly robust in such distorted speech, it is also shown that detailed spectral features are better able to model listeners’ responses to undistorted speech. [Work supported by SSHRC.] a)Now at the Dept. of Psychology, University of Wisconsin, Madison, WI.