Bi-microphone speech enhancement

Abstract

Speech enhancement is an operation that aims to reduce acoustic noise picked up by noisy recordings of speech. It is essential to various systems such as mobile communications, hearing aids and hands free car telephony where the speaker's voice is often sampled in adverse acoustical conditions. Various solutions and implementations already exist to enhance the intelligibility and quality of speech in noisy environments and can generally be classified in two different groups; single microphone and multi-microphone methods. Single microphone techniques rely on the statistical properties of speech and noise to isolate one from another while multi-microphone techniques generally exploit spatial characteristics of the speaker and noise sources, often requiring precise microphone placement and calibration. This thesis proposes a novel and robust bi-microphone system aimed to fill the gap between the two extremes. Speech coherence between the two sound sources is exploited to supplement state of the art single microphone methods that use suppression rules and effective noise estimation techniques; the requirements on microphone directivity and placement are minimal, which enables rapid integration in current mobile devices. Subjective and objective results attest of the system's improved noise suppression compared to state of the art techniques in low signal to noise ratios.