A new robust adaptive beamformer for enhancing speech corrupted with colored noise

In this paper, a new adaptive beamforming scheme is proposed for enhancing a speech signal degraded by additive colored noise signals in reverberant environment. The good performance of the standard GSC beamformer is severely impaired in complicated acoustic environments. The proposed beamformer uses a new adaptive blocking matrix, which consists of linear prediction error filters (LPEFs) and a multi-channel noise canceller using adaptive noise estimation filters (ANEFs). The new adaptive blocking matrix exploits the stationary characteristics of the speech signal over a short interval of time. Thus most of speech signal can be predicted by the linear prediction filters in the blocking matrix. On the other hand, the background noise signals becomes white. While the ANEFs in the multi-channel noise canceller reconstructs the background noise signals from whitened noise signals. Simulation results, in a noisy and reverberant environment with 0.3s reverberation time, show that the proposed beamformer produces results that are significantly favorable than standard GSC beamformer. Moreover, the proposed beamformer has been implemented in a real-time in a real environment using National Instruments NI-PXI-1042Q controller system and data acquisition card PXI-4472.