Cascaded Structure of Noise Reduction and Multiband Frequency Compression of speech signal to Improve Speech Perception for monaural Hearing Aids
Rajani S. Pujar, Pandurangarao N. Kulkarni · 2019
This paper presents a cascaded structure of noise reduction and multiband frequency compression in enhancing speech perception for sensorineural hearing impaired persons using monaural hearing aid. In multiband frequency compression scheme, input signal is divided into 18 frequency bands ranging from 0 to 5 KHz based on auditory critical bandwidths (ACB). Frequency components of each band are compressed at the center of their respective bands using frequency segment mapping method. This results in concentration of all spectral energy at the center of each band to minimize the effect of spectral masking. Noise reduction scheme, namely wiener filter is used to reduce the background noise. Evaluation process involves i) Speech quality assessment (using PESQ and MOS) on normal-hearing subjects ii) Speech intelligibility assessment (using MRT) on hearing-impaired subjects. The test material consists 50 sets of consonant-vowel-consonant (CVC) words. Each set includes six words. Each subject is responded for 1800 presentations (300 CVC words x 6 different SNR conditions). The MOS, PESQ and MRT scores obtained for the SNR values ranging from -6 dB to + 6 dB indicates enhancement in speech quality and intelligibility.