Acoustics, Speech, and Signal Processing, 2004. Proceedings. (ICASSP '04). IEEE International Conference on
Martin Fuchs, Gerhard Schmidt
In noise suppression systems for automotive applications the useof adaptive beamformers has proven to be of great potential. Nevertheless,in diffuse noise fields the amount of noise attenuation israther limited and depends on the number of microphones. In orderto enhance the signal-to-noise ratio further, additional classicalnoise suppression schemes, like spectral subtraction, are often applied.Unfortunately, these schemes tend to introduce either speechdistortions or leave a large amount of residual noise. In this paperwe describe a method of extracting additional spatial informationfrom a conventional beamformer in generalized sidelobe structure.This spatial information can be utilized, e.g., to control parameterslike overestimation or spectral floor of classical noise suppressionschemes in a frequency selective manner or to compute a simpleattenuation factor for suppressing nonstationary noise. An outlookis given on further usage of the spatial information in other algorithmicparts of a hands-free telephone or a speech recognitionsystem.