Subjective speech-to-noise-ratio as a single absolute performance measure for digital waveform coders
Mamoru Nakatsui · The Journal of the Acoustical Society of America · 1980
The ultimate performance measure for evaluating a voice communication system is the subjective quality of the received speech. To this end we tested subjective speech-to-noise-ratio (SNRsubj) as an absolute measure of overall speech quality for digital waveform coders. High quality digital recordings of two sentences by each of two speakers were processed by nine different waveform coder and transmission rate combinations, and then paired with reference signals corrupted by varying amount of multiplicative white noise. The SNRsubj of a coded test signal is defined as the amount of multiplicative white noise introduced into the uncoded original signal so that it is judged equally preferable to the coded signal. A total of 392 pairs were randomized and presented to 11 naive listeners for preference testing. The SNRsubj for the nine coder configurations varied from 14 to 40 dB and was found suitable to represent overall speech quality of waveform coders in a single dimension. None of two main experimental factors, speech sample and listener, introduced any statistically significant variation in the SNRsubj estimate. The relationship between the SNRsubj and several objective measures are also discussed.