conference paper

Conference/ProceedingsElektronische Sprachsignalverarbeitung ESSV 2004
Start date20.09.2004
End date22.09.2004
Author(s)Peter Noll, Markus Schwab, Wiryadi
TitleSensing People - Localization with Microphone Arrays
AbstractIn this paper we present a real-time microphone array system which performs 3D source localization, multi channel speech enhancement and robust speech recognition. The acoustic source localization uses the SRP-PHAT method [1] to produce
potential source locations. A clustering algorithm excludes outliers and enables a multi source tracking. The localizations are finally optimally filtered with an appropriate Kalman filter. The proposed speech enhancement, a weighted subarray Delay-and-Sum beamformer, is designed to cope with the problem of diffuse noise and changing speaker positions subject to minimization of the word error rate (WER) of an automatic speech recognition system (ASR). The proposed algorithm reduces the WER by more than 50 % compared to the WER of a single microphone signal.