필터뱅크 기반의 프로스트 빔포밍 최적화

Abstract

Beamforming is intelligent spatial filtering which extracts only target signals from noisy environment using microphone arrays. Fixed beamforming is simple and easy concept but does not show good performance in real noisy situations. As an adaptive method, Frost beamforming which has error correction feature is a good alternative. Furthermore, to adjust elements of each filterbank such as filter length and learning rate, Quadrature Mirror Filter (QMF) is implemented. The filter lengths and learning rates of the beamformer are optimized in each band. The performance is evaluated by Signal to Noise Ratio (SNR) gain and Bark scale Spectral Distortion (BSD).