Subjective Evaluation of PCM Coded Speech

"How does it sound?" This is a fundamental but elusive question for the engineer designing or evaluating a system for transmitting, recording, or processing speech signals. If the system is analog, the engineer has as a guide a substantial body of information about the interrelated effects on speech quality of such factors as attenuation, noise, linear and nonlinear distortion, echo, and cross-talk. 1 With respect to digital systems, however, the subjective effects of characteristic distortions have been documented to a much smaller extent and, as a consequence, the quality of an existing system and the merits of proposed designs are much harder to predict. One approach to the evaluaton of digital systems is to relate a digital signal distortion to one of the analog distortions, and to define digital speech quality as the subjective correlate of the equivalent analog distortion. 2 Although expedient and reasonably accurate for 1087 certain individual distortions, the value of this approach seems quite limited in the important situations where several distortions occur simultaneously. While the engineering literature contains many reports of subjective tests of digitally coded speech, most of the tests were undertaken to provide performance data on the overall distortions produced by specific coders. Among the exceptions to this approach and more aligned with the spirit of our work are the experiments reported by Donaldson and Chan, 3 O'Neal and Stroh, 4 and Yan and Donaldson 6 in which individual sources of distortion were identified and the manner of their interaction investigated.