Perceptual Multiband Spectral Subtraction for Noise Reduction in Hearing Aids

Abstract

This paper presents a low complexity noise reduction method by a perceptual multiband spectral subtraction for CIC type hearing aids. The noisy speech is first perceptually decomposed by a 1/3 octave filter bank by ANSI S1.11, and the noise is estimated by an enhanced entropy based voice activity detection (VAD), and then noise is subtracted by a simple multiband spectral subtraction. Simulations show that under different environments the proposed algorithm averagely improves segmental SNR (Signal to Noise Ratio) and PESQ (Perceptual Evaluation of Speech Quality) by 6.27dB and 0.32, respectively, and performs better compared to the previous approach. In addition, informal listening test shows there is not musical noise in the processed speech.