Performance Evaluation Of Deconvolved Beamforming For Localization Of Distance Speech Signals In Outdoor Environment

Speech perception is an important content of acoustic signal processing.In outdoor environments,distant speech signals are always polluted by the environmental noise and various interference sources.Enhancement of distant speech is a challenging task in microphone array processing.The dissertation focuses on high-resolution estimation of direction of arrival(DOA)of speech and the wideband constant beamwidth beamfoming teclhnique for speech enhancement.With less snapshot data,the deconvolved conventional beamforming(DCBF)implemented to a small size array can achieve super-directionality as high-resolution beamforming,and at the same time retains the robustness of conventional beamforming.In this dissertation,the existing one-dimensional DCBF is extended to two-dimensional DCBF(2D-DCBF),by which the azimuth and elevation of speech signals can be estimated at the same time.Further,the Cramer-Rao lower bound is derived in the u-space,and the Monte Carlo method is utilized to analyze the validity of the 1D and 2D-DCBF algorithms.In the framework of incoherent signal-subspace method,the DCBF algorithm is applied to the estimation of the DOA of the broadband signals.Especially,the DCBF algorithm provided is suitable to arbitrary shape microphone arrays.The outdoor experimental data processing results have shown that the 2D-DCBF implemented to a microphone array can effectively estimate DOA of distant speech signals.Due to the speech band covering 5 octaves,the classical constant beamwidth beamforming algorithm in the frequency domain is not suitable for enhancement of distant speech.This dissertation combines the spatial response constraint(SRV)and the sidelobe control(SL)together as the cost function,and uses the convex optimization approach to solve the beam pattern synthesis problem to achieve a frequency-invariant response.To improve the precision of beam direction,the upper limit constraint of beam response is imposed.Moreover,implementation of Tapped delay lines(TDL)to integer pre-delays in advance is combined with the finite impulse response(FIR)filter with real coefficients for designing constant beamwidth beamforming in the time domain with a low computational complexity.Numerical simulation shows that the proposed constant beamwidth beamforming algorithm can satisfy the requirement of speech enhancement.The experimental data processing results have validated the effectiveness of the proposed speech enhancement processing framework.