Minimum phase whitening filter to speech beamforming technique

Abstract

This paper presents application of minimum phase whitening filter for speech enhancement technique to two-microphone beamforming structure. The method uses switching technique for updating the estimate alternately between frontend minimum phase whitening filter during speech plus noise period and rear-end adaptive finite impulse response (FIR) filter during noise alone period. For application of speech beamforming technique, innovations-based minimum phase whitening filter is applied in front of sum and subtract (SS) function with the use of one sample delay filter in rear-end primary input. A test result shows that the proposed approach by alternately switching minimum phase whitening filter and adaptive FIR filter provides an improved performance of 5.9dB and 18.03dB in signal-to-noise ratio (SNR) in stationary noise and non-stationary noise, respectively than adaptive Griffiths and Jim (G-J) beamformer by alternately switching two adaptive FIR filters.