Near-field Adaptive Beamformer for Robust Speech Recognition

This paper investigates a new microphone array processing technique specifically for the purpose of speech enhancement and recognition. The main objective of the proposed technique is to improve the low frequency directivity of a conventional adaptive beamformer, as low frequency performance is critical in speech processing applications. The proposed technique, termed near-field adaptive beamforming (NFAB), is implemented using the standard generalised sidelobe canceler (GSC) system structure, where a near-field superdirective (NFSD) beamformer is used as the fixed upper-path beamformer to improve the low frequency performance. In addition, to minimise signal leakage into the adaptive noise canceling path for nearfield sources, a compensation unit is introduced prior to the blocking matrix. The advantage of the technique is verified by comparing the directivity patterns with those of conventional filter-sum, NFSD and GSC systems. In speech enhancement and recognition experiments, the proposed technique out-performs the standard techniques for a near-field source in adverse noise conditions.