Speech enhancement using resonator filterbanks
Luc Gagnon, W.F. McGee · 1991
A speech enhancement algorithm for wideband noise reduction based on an analysis-synthesis resonator filter bank (ASRF) configuration is proposed. This choice is motivated by the observation that the configuration demonstrates excellent synthesis properties when time-varying spectral amplitude modifications are allowed at the output of the analysis stage and because the resonator filter banks are computationally efficient when implemented on single-chip digital signal processors. The noise reduction technique is in the family of algorithms which perform spectral decomposition of the noisy signal to estimate the short-time spectral amplitude of the speech signal. The ASRF speech enhancement algorithm was tested on speech degraded by different kinds of broadband noise and the enhanced speech was compared with material produced by other speech enhancement algorithms.>