A Robust Speech Enhancement Scheme on The Basis of Bone-conductive Microphones

Abstract

This paper presents a novel speech enhancement scheme on the basis of bone-conductive microphones. High performance in non-stationary noisy environments is achieved by combining the following two techniques: the speech detection based on the bone-conductive microphone signal, and non-stationary noise suppression by adaptive spectral subtraction. Unlike regular air-conductive microphones, bone-conductive microphones are insensitive to environmental noises and background speech. By employing this advantage, we are able to detect very robustly speech in non-stationary noisy environments. Furthermore, based on the reliable speech detection results, the adaptive spectral subtraction can operate effectively even under low SNR conditions.