Study On Speech Enhancement Adaptive Beamforming Algorithm Based On Generalized Sidelobe Canceller

IntroductionSpeech enhancement technique has been applied in many areas of speechsignal processing, such as noise canceling, speech compression, speech encodingand recognition etc., since in practical speech communications, speech signal hasalways been disturbed by environmental noises. These noises disturb speech signalin different extent and make the performance of the speech processing systemsworse.Speech enhancement algorithms based on multi-channel input use not onlyfrequency information, but also spatial information. Compared with speechenhancement algorithms based on single-channel input, they can improveperformance index of speech enhancement and could deal with more kinds of noise.Therefore they get more and more attention by both domestic researchers andresearchers abroad. Many researchers have conducted studies on speechenhancement algorithms on the basis of beamforming. In most studies, this kind ofalgorithms works well only in the directional point noisy environment. But in thediffused noisy environment, the performance of these algorithms is degraded. Onthe other hand, a large number of sensors used in this algorithm would make thesystem operation complexity larger. This is not practical and suitable for the realapplications. How to improve the algorithm performance and in the same timemake it more practical is a problem that need more study.After the above consideration and analysis, this paper provides an adaptivebeamforming speech enhancement algorithm based on GSC (Generalized SidelobeCanceller). This algorithm utilizes beamforming algorithm with GSC structure,subband structure, frequency subband decomposition technique and partiallyadaptive technique, to improves noise canceling performance, convergent rate, lessalgorithm complexity. In addition, it also makes use of speech nonstationaritycharacteristics to estimate transfer function between signal source and sensors,which can make the algorithm better in practical implementation.1. Beamforming Algorithm based on GSC StructureBeamforming technique belongs to spatial filtering. It can separate signals thathave overlapping frequency content but originate from different spatial locations.In the conditions that the desired signal is disturbed by noise and other interferingsignal, to hold the signal from desired direction and depress the noise and otherinterfering signal using a beam with some certain form can be obtained. Beamformers can be classified as data independent beamformers, statisticallyoptimum beamformers, adaptive beamformers and partially adaptive beamformers.Each kind of beamformers has different noise canceling performance in differentnoisy environments. GSC is a statistically optimum beamformer. Throughthree-part structure, it can change a linear constrained problem in LCMV (LinearlyConstrained Minimum Variance) beamformers into unconstrained form. Thismakes GSC has a lot of applications in practice. The GSC algorithm is comprised of three components: (1) Fixed Beamformer (FBF) Fixed beamformer is actually a simple fixed filter. It simply processes noisysignal. The output signal from fixed beamformer contains a majority of the desiredsignal and a small part of noise signal. (2) Blocking Matrix (BM) Blocking matrix is used for generating reference noise signal. After the noisysignal from sensors enters the blocking matrix, the desired signal is cancelled andthe noise signal is kept, so that the output of the blocking matrix is just noisesignal. (3) Noise Canceler (NC) Reference noise signal enters noise canceller. By adjusting the weight factors,the noise signal could be canceled clearly and effectively using the output of noisecanceller and the output of fixed beamformer.2. Study on Speech Enhancement Adaptive Beamforming Algorithm Based on GSC In this paper, the proposed algorithm uses GSC beamforming method, subbanddecomposition technique and partially adaptive transformation technique toimproves performance of the adaptive algorithm and obtain a better speechenhancement algorithm. Transformation Matrix T Signal prior information for the realization of the matrix transformation of the V...