Abstract: A novel method using signal-adapted filter banks to carry out the noise reduction is investigated. Since signal-adapted filter banks match the statistics of the underlying signal, more efficient signal decomposition can be obtained. The filter banks are designed based on teh low pass subband energy maximum according to the statistics of the signal. Combining signal-adapted filter banks and the traditional thresholding technique, an efficient noise reduction method is presented. Experimental results show that the novel method achieves larger improvement of SNR than the standard filter banks methods at the same filter length.

Key words: signal-adapted filter bank, linear semi-infinite programming, hard thresholding, noise reduction