Abstracts Submitted: 510
Number of Users: 708
View Abstracts Submitted
Back to home Page
Beamforming, that is, spatial filtering, has been one of the important issues for noise reduction. The authors have proposed a neural network-based broadband beamformer, which forms the sharp main lobe and decreases side lobes. There is, however, the tradeoff between sharpening the main lobe and the degree of the signal distortion. It is necessary for speech applications to achieve the wideband optimization of the beampattern. In this paper, the beampattern optimization is achieved using dual cost functions, which rely on the beam pattern and the spectral distortion, respectively. The dual cost functions contribute to achieving the sharp main lobe with less distortion on the target signal. The feasibility of the proposed method is confirmed by computer simulation.
© Copyright 2017 All Rights Reserved