Speech Enhancement for Mobile Phones Based on the Imparity of Two-Microphone Signals

Abstract

In this paper, we consider a speech enhancement system for mobile phones using pair microphones. The locations of the two microphones are arranged to facilitate the derivation of speech absence probability. The results are substituted into an MMSE estimator in loudness domain for speech purification. Compared with the conventional algorithms [1, 2], which estimate the speech absence probability based on a single microphone, the proposed method does not require many empirical parameters and is relatively robust under different types of noises. The performance of this method is evaluated under noisy environments with music and car noises.