Frequency compression of speech for improving speech perception in sensorineural hearing loss: FBS approach

Abstract

Filter-bank summation (FBS) method is a commonly used technique for multi-band processing of speech and audio signals, especially in digital hearing aids. In multi-band speech processing techniques, filter bank summation provides the convenient way of processing the auditory information present in different bands based on their perceptual significance. People with hearing loss have issue in perception of speech due to widening of the auditory filter bandwidth leading to increased frequency masking. Previous studies have shown that spectral splitting of speech signal for binaural dichotic presentation helps to reduce the effect of frequency masking. Also studies showed that using multiband frequency compression it is possible to compensate the effect of widened auditory filters. This paper presents a filter bank summation method to perform dichotic spectral splitting of input speech signal followed by frequency compression to enhance speech perception for hearing impaired. In the present study, the speech signal is split into eighteen frequency bands ranging from 0–5000 Hz based on auditory critical bandwidths and frequency samples of every band compressed in the direction of center of each band using spectral segment frequency mapping technique. Performance of the algorithm was evaluated using MOS test for subjective assessment of speech quality and Perceptual Evaluation of Speech Quality (PESQ) scores for objective assessment of speech quality. The results showed a significant improvement in speech quality as indicated by MOS and PESQ scores for the SNR values in the range of −6 dB to 6 dB.