Noise reduction and loudness compression in a wavelet modelling of the auditory system

Describes the modelling of the normal auditory and hearing-impaired system for hearing aid design and addresses two major problems: noise reduction and loudness compression. This new integrated approach is based on a wavelet transform and a formulation of temporal and spectral psychoacoustic models of masking and loudness sensation. Within the model, a new subtraction algorithm is developed to model the masking phenomena and reduce noise in single-input systems. As a result, the enhanced speech quality has been significantly improved, especially in unvoiced portions. The noise component during silent periods has been attenuated by up to 20 dB. The auditory loudness sensation is also modelled with a homomorphic AGC for time-varying frequency-dependent compensation for loudness recruitment. The dynamic nonlinear weighting is a function of the wavelet coefficients. The homomorphic approach is found to be superior to linear AGC. Finally, the performance of the whole model is evaluated and compared with that obtained using the Chabries's model, and a significant improvement is found. It is concluded that optimal wavelet subtraction and homomorphic AGC are an appropriate combination to synergise noise reduction and loudness recruitment problems in hearing aid design.