Magnitude estimation of degraded speech quality by normal- and impaired-hearing listeners

Speech quality magnitude estimates (SQME's) were obtained from 12 sensorineurally impaired-hearing listeners on connected speech samples degraded by changing low-pass filter cutoff frequency, high-pass filter cutoff frequency, or percent total harmonic distortion (THD) by linear rectification. Log SQME's varied linearly with log bandwidth for filtered signals and with log percent undegraded (100-% THD) for linearly rectified signals. Significant variations were found among the slopes of the log-log functions for degradation modes and group-by-degradation mode interactions. Slope differences appeared to represent differential sensitivity of the listeners to changes in mode and degree of degradation. The findings are sufficiently encouraging to suggest that direct scaling procedures be employed in future studies of the evaluation of communication systems and the perception of complex signals.