Perceptual and Objective Evaluation of Speech Processed by Adaptive Differential PCM

Speech signal processing systems are susceptible to a variety of audible impairments often classified with words like "distortion," "noise," "echo," and "sidetone." These categories are themselves subdivided: for example, "linear" and "nonlinear" distortion, "white" noise, "impulsive" noise, "speech-dependent" noise, etc. When the type of system is familiar to a large body of listeners, the application of these names becomes standardized and a language exists for describing the quality of specific implementations. With new systems, however, the types of degradation are often not known a priori, and special effort is required to identify them and to relate them to physical characteristics of the system. For example, experiments on PCM (pulse code modulation) have identified peak clipping, granular quantizing noise, and bandlimiting as important audible degradations.1-2 In PCM, there are relatively few design parameters, and each of these impairments can be related to one of them: peak clipping to quantizer overload point, granular noise to step size, and bandlimiting to sampling rate. 1597