Quantitative and qualitative (subjective) perceptual measures for speech processing applications

This paper addresses perceptual measures, both quantitative and qualitative (subjective) that are indicative of the effectiveness of the speech processing performed by a digital hearing aid. The system (based on Texas Instruments TMS320C3X digital signal processing (DSP) chip) can implement spectral shaping, noise reduction (speech enhancement), multichannel compression and multichannel interaural delay. This paper presents three quantitative measures of evaluating the processed signal, using the adaptive noise reduction option, and correlates these measures with qualitative feedback obtained from human subjects. The three quantitative measures discussed are signal to noise ratio (SNR) improvement, relative signal power using the power density spectrum, and spectral distance measured by root mean square (RMS) logarithm spectral distortion. For the subjective measure, an experiment was designed using a test data set consisting of eight sentences.