Abstract:

Speech is the most efficient auditory code for conveying abstract information. Hence, high-fidelity speech transmission is an important goal for some multimedia applications. Problems may arise when the signal processing used to create the artificial auditory display inadvertently alters the spectral/temporal properties of the speech or when competing signals mask out portions of the speech. In cases where the distortion amounts to linear filtering and/or noise masking, the degrading effect on intelligibility can be predicted using an elaborate calculation scheme known as articulation theory [H. Fletcher and R. H. Galt, J. Acoust. Soc. Am. 22, 89--151 (1950)]. The theory assumes that the audible spectrum can be divided into a number of bands that contribute independently to speech reception; and the band contributions are additive. Acoustical measurements are translated into their perceptual correlates via transformations derived from psychophysical experiments. These terms are combined to give the articulation index a number which is directly related to the expected percent-correct score for an articulation test. The theory will be described and examples presented. [Work supported by NIH R29 DC02127.]