Post-Filtering Techniques For Directive Non-Stationary Source Combined With Stationary Noise Utilizing Spatial Spectral Processing

This paper investigates the problem of enhancing a desired speech source corrupted by a stationary noise. A beamformer structure is proposed which combines an adaptive constrained beamformer with a post-filtering technique. The adaptive constrained beamformer is designed to spatially enhance the desired source from the observed signal. After the adaptive constrained beamformer, a post-filter is employed to increase noise suppression capability. Experimental results in a real car situation show that the proposed structure achieves a good noise suppression level with a low distortion