Kepstrum approach to real-time speech-enhancement methods using two microphones

Abstract

The objective of this paper is to provide improved real-time noise canceling performance by using kepstrum analysis. The method is applied to typically existing two-microphone approaches using modified adaptive noise canceling and speech beamforming methods. It will be shown that the kepstrum approach gives an improved effect for optimally enhancing a speech signal in the primary input when it is applied to the front-end of a beamformer or speech directivity system. As a result, enhanced performance in the form of an improved noise reduction ratio with highly reduced adaptive filter size can be achieved. Experiments according to 20cm broadside microphone configuration are implemented in real-time in a real environment, which is a typical indoor office with a moderate reverberation condition.