Kepstrum analysis and real-time application to noise cancellation

Abstract

This paper provides analysis of kepstrum (known elsewhere as complex cepstrum) and its real-time application to noise cancellation. The method uses kepstrum estimation for identification of the ratio of acoustic path transfer functions between two microphones and efficient processing technique for real-time processing. It is applied to the front-end of speech enhancement method, the Griffiths and Jim beamformer. The experiments are processed in real-time by software implementation in LabVIEW. It will be shown that the front-end kepstrum approach effectively cancels echoes and hence reverberation in a noisy environment. For a real-time application, the computational complexity on kepstrum processing has also been investigated and compared with NLMS algorithm of adaptive filter.