DOA Estimation in Car for Abnormal Sound Localisation

Abstract

Spurious noise in car cabinet can be not only annoying bust also indicative of some
potential faults. A small square microphone array with 4 sensors was adopted in this paper to
localize the sound source in car for fault diagnosis. A new voice activity detection (VAD) algorithm
was proposed for the typical discontinuous short-time noise in car due to some fault and applied to
direction of arrival (DOA) estimation as a pre-processing stage. Four different time delay estimation
methods were compared based on the measurements from a typical passenger. Experimental results
illustrate that the VAD algorithm is crucial to achieve robust fault localization performance and the
generalized cross-correlation method with phase transform weighting function is an appropriate
fault localizer in car.

Document Downloads