Perception of digitally coded speech under some cognitive load

Abstract

Perception of meaningful sentences that were encoded by speech coding methods was measured. In order to differentiate and to expand perceptual differences between the speech coding methods for the meaningful sentences, a task of recall for the sentences under cognitive load was adopted. In this experiment, the distractor method of short-term memory [L. R. Peterson and M. J. Peterson, J. Exp. Psychol. 58, 193–198 (1959)] was utilized as the cognitive load to subjects. Pitch predictive multipulse speech coding at 8 kb/s [K. Ozawa and T. Araseki, Proc. ICASSP, 1689–1692 (1986)] and 50 kb/s μ-law PCM were used as speech coding methods. As a reference, uncoded natural speech was also used. A total of 60 Harvard sentences was used as meaningful sentence stimuli. Results showed that the higher the cognitive load, the larger the perceptual differences between uncoded and coded speech, as well as between MPC and PCM. However, differences in error rates among uncoded and coded speech were very small in the case of no cognitive load. The results suggest that this perceptual evaluation method under cognitive load may be a sensitive method that can differentiate perceptual differences in speech quality between coded speech, even for meaningful sentences.