An optimized fixed equalizer for speech enhancement

Abstract

This paper proposes a new method to design the fixed single-channel equalizer parameters based on perceptual evaluation of speech quality (PESQ) and the short-time objective intelligibility (STOI). The proposed equalizer does not require the need for noise estimation nor a voice activity detector. The fixed weightings are obtained by a training process containing a set of training data. A multi-objective optimization problem was formulated with the PESQ and STOI as the objective measures, which results in a set of candidate solutions via the Pareto optimality. Experimental results demonstrate improvements in PESQ and STOI, and the trade-off between these two measures is well defined in a set of Pareto optimal solutions to allow flexible adjustment.