In hands-free audio conferencing applications, microphone arrays [1][2] using simple beamforming technique often cannot differentiate between the scenarios where (I) a single talker speaking in the presence of acoustic reflections, and (II) two individual talkers having a conversation. Various approaches have been proposed to solve this problem [1]-[6]. However, most of these approaches are either beamforming technique specific or computational demanding. In this study, we proposed a new method that uses the maximum correlation lag to distinguish a new talker from acoustic reflections. The proposed method considers the power of beamformer outputs instead of raw microphone signals. Therefore, the proposed method allows any beamforming technique to be employed, and at the same time, lowers the computational requirement making a real-time implementation possible. Experimental results performed in an anechoic chamber and in a reverberant room show that scenario I always has a greater amount of maximum correlation lag, whereas scenario II always results in less correlation lag. These results suggest that correlation lag could be used as an effective means for distinguishing talkers from reflections.

, , ,
IMTC'05 - Proceedings of the IEEE Instrumentation and Measurement Technology Conference
Department of Systems and Computer Engineering

Lo, D. (David), Goubran, R, & Dansereau, R. (2005). Acoustic reflections detection for microphone array applications. In Conference Record - IEEE Instrumentation and Measurement Technology Conference (pp. 1139–1143). doi:10.1109/IMTC.2002.1007117