REAL-TIME IMPLEMENTATION OF SPECTRAL SUBTRACTION MODIFIED BY WIENER FILTERING

摘要

Spectral subtraction is very popular method for noise suppression in speech signals due its robustness and the simplicity for implementation. The main shortcoming of spectral subtraction is the necessity to estimate noise characteristics during speech pauses. Any voice activity detector (VAD) is by this methods mostly required. This paper describes an approach of noise suppression based on the combination of spectral subtraction with iterative Wiener filtering. The main advantage of this method is the possibility to estimate a background noise even if the speech signal is presented. This method does not require any VAD. The extended spectral subtraction was implemented in real-time on fixed-point DSP.