Wavelet Filter Banks in Speech Enhancement Algorithms

V. Davídek (Czech Republic)


speech enhancement, wavelets, subband processing, filter banks.


Speech enhancement systems are developed to work with one sampling frequency as full band systems or as a multirate systems splitting the input signal into subbands. This paper deals with the problems of subband processing techniques used in speech enhancement algorithms. Four basic algorithms of speech enhancement are compared in full band processing and subband processing. Speech signals corrupted by an additive white noise and a running car noise are used for testing of adaptive Wiener filtering, spectral subtraction, coherence method and extended spectral subtraction. For subband processing is used the wavelet octave filter bank based on the optimized FIR filters.

Important Links:

Go Back