Intelligibility of speech processed by a spectral contrast enhancement procedure and a binaural procedure

Abstract

The study reported was undertaken as an attempt to improve the intelligibility of selected speech signals (numerical test) masked by a speech-shaped noise, by a proposed algorithm of the speech signal processing based on the spectral contrast enhancement and binaural procedure. The spectral contrast enhancement involved the transformation of the speech signal spectrum to achieve an increase in the level of the amplitude between subsequent minima and maxima of individual formants. The binaural procedure involved the determination of the interaural time difference (ITD) and the interaural intensity difference (IID) in order to select and enhance the fragments reaching the listener from the front. The intelligibility of speech processed by the algorithm was compared to that of the unprocessed speech for different spatial separations between the signal and the noise sources and for a few signal to noise ratios. The results have shown that the intelligibility of the speech presented against the noise significantly depends on the angle between the directions of the signal and the noise. The intelligibility of speech is much larger when the noise and the speech reach the observer from different directions, i.e. when their sources are spatially separated. In general, the algorithm improves the intelligibility of numerical tests presented at the background of the noise by a few percent and the greatest improvement has been observed for low signal to noise ratios. The algorithm performance was different for each hearing impaired subject with binaural hearing loss. To get an objective assessment of the algorithm performance it should be optimised and tested on a larger group of subjects for more diverse sound material.