Intelligibility and ratings of digitally coded speech

Abstract

An experiment has been performed to investigate perceptual effects of digital encoding of speech. The aim was to gain information about the influence on intelligibility and subjective quality of three distortions: bandwidth reduction, peak clipping, and amplitude quantization. Allowing two levels of each distortion, we produced eight encoders encompassing all possible combinations of the three impairments. By means of a consonant recognition test (CRT) and a category-rating test, twelve subjects provided intelligibility and rating data about each encoder. The results show that the effect of multiple distortions is not, in general, the sum of effects of individual distortions and that the distortions influence intelligibility and subjective quality differently. For example, with respect to consonant recognition, quantization and clipping reinforce one another; occurring together they cause more recognition errors than the sum of the errors caused by the two distortions occurring individually. On the other hand, the effects of quantization and clipping on subjective quality are essentially additive.