Incorporation of temporal masking effects into bark spectral distortion measure

The objective of this paper is to extend a promising objective speech distortion measurement method, the bark spectral distance (BSD) measure, with the auditory concepts of forward and backward temporal masking to improve its measurement accuracy. The results of this investigation show that automatic BSD-based speech quality ratings may be made to correlate better with existing MOS ratings by removing perceptually irrelevant areas of speech from the distance measure. The correlation between the objective BSD measure to the subjective MOS measure increases from 0.91 to 0.98. The best results were found with a window duration of 128 samples, use of exponential-slope filter characteristics for both forward and backward masking effects, forward masking delays up to 100 msec, and a backward masking time advance of 40 msec