A subband hybrid beamforming for in-car speech enhancement

Abstract

In this paper, a multichannel speech enhancement system is presented, dedicated to in-car communication. An experimental study of the acoustic field inside the car interior leads us to propose a hybrid beamforming algorithm, taking two frequency ranges into account, according to the multichannel noise signal coherence. While in the high frequency range, the noise signals for different microphones show little coherence, the same signals are strongly coherent in the low frequency range. This observation leads us to design a hybrid system where an adaptive Minimum Variance Distortionless Response (MVDR) beamforming is used in the high frequency range, while a Speech Distortion Weighted Multichannel Wiener Filter (SDW-MWF) is applied in the low frequency range. The performance of the algorithm is evaluated by objective performance measurements on real in-car audio data, and shows promising results.