Interaction of speech disorders with speech coders: effects on speech intelligibility

Modern speech coding schemes have been developed to address the demand for economical spoken language telecommunication of acceptable quality. A variety of speech coding algorithms have been described, which compress speech to facilitate efficient transmission of spoken language over communication networks ((J.R. Deller Jr., 1993; P.E. Papamichalis, 1987). Most such speech coding algorithms are lossy in the sense that the processed speech is not identical to the original speech. As a result, some distortion is invariably introduced with any lossy speech coding strategy. For this reason, candidate coders undergo detailed evaluation to ensure that the associated speech output is of acceptable quality (S.R. Quackenbush et al., 1988). Three different coding algorithms were investigated relative to unprocessed speech: the Codebook Excited Linear Prediction (CELP), the Global System for Mobile Communications (GSM) algorithm which is a standardized speech coding algorithm in Europe, and the Linear Predictive Coding (LPC) algorithm. The specific coding schemes evaluated were MatLab implementations of NSA FS-1015 LPC-l0e; NSA FS-1016 CELP-v3.2; and ETSI GSM (A. Spanias, 1995). One of the goals of this study was to quantify the coding distortion using objective measures and to correlate these measures with speech intelligibility and subjective quality data, in the hope of identifying one or more measures that can predict the subjective results.