Speech quality assessment using expert pattern recognition

Abstract

A technique is described for processing objective parameters to obtain voice quality assessment. The method combines statistical pattern recognition (SPR) techniques with a knowledge-based expert system. The SPR module uses optimally selected voice parameters to form a purely statistical assessment of voice quality. Output includes a classification of the most likely opinion score, a prediction of mean opinion score, and the predicted opinion score frequencies. An expert system is being developed to combine these results with additional knowledge bases to account for factors, such as delay, that are not accounted for by the SPR system. >