A dual-channel beamformer based on time-delay compensation estimator and shifted PCA for speech enhancement

Speech enhancement is an essential technique to process degraded audio in various applications. Beamforming to eliminate interferences based on sensor arrays is the most well-known method for this issue. However, traditional beamformers often face magnitude incoherence towards received signals due to directional weighting. Therefore, a novel dual-channel beamformer based on time-delay compensation (TDC) and shifted principal components analysis (PCA) is presented in this work. Firstly, our enhancement algorithm utilizes TDC estimator to preserve binaural cues, including interaural time-delay and intensity difference. Then the estimated cues are comprised to improve the shifted PCA, which can reduce noise by extracting primary components. Finally, the aforehand processed audio are input to a beamformer with post-filter to obtain enhanced speech. Experiments have demonstrated that the proposed algorithm could achieve some superiorities in speech intelligibility compared with the state-of-the-arts against real scenarios.