Noise suppression using one-regular Unimodular filterbanks

In this paper one-regular unimodular filterbank (UMFB) is used for subband based noise reduction in speech signals. The regularity property of the filterbank enables a better estimate of the noise characteristics at low frequencies. This results in effective reduction of noise when compared with conventional filterbanks. Unimodular filterbanks have the lowest possible system delay of M - 1 samples, where M is the number of channels. Hence, we show that effective noise reduction in speech signals can be obtained, with UMFB, at a very low system delay. The implementation of UMFB is done with simple structures having low computational complexity. The results are validated with suitable experiments.