Subjective evaluation of pcm coded speech

Abstract

Subjective quality ratings of pcm coded speech were obtained with the aims of (i) determining the effects of certain coder parameters and their interactions on speech quality, (ii) finding objective measures for predicting perceived distortions, and (iii) providing guidelines for optimizing coder design. Coders with various combinations of four clipping levels, seven step sizes, four bandwidths, and three logarithmic companding laws were simulated. The coders were rated for quality on a 10-point scale by 48 listeners who heard male and female speech processed by the coders. The ratings depended strongly on clipping level and step size, but only weakly on bandwidth. None of the coder parameters interacted strongly with another. Clipping noise power grossly overestimated the extent of perceived overload distortion; instead, clipping percentage is proposed as a much more realistic predictor. Signal-to-granular-noise ratio was a good predictor of perceived granular noise. For a given bit rate, the coder with the highest quality rating was not the coder with minimum total clipping and granular noise power, contrary to traditional wisdom.