Consideration of noise substraction coefficient in noise reduction system based on modified DFT pair

摘要

We have proposed a noise reduction system based on the modified DFT pair, where the input and the noise spectra were detected in real-time. The noise reduction was achieved by subtracting the detected noise spectrum from the input one. The noise subtraction coefficient was also introduced for adjusting amount of the noise spectrum to obtain the optimal output SNR and was determined by trial and error. However, such the optimal value did not bring the best performance in the subjective evaluation. In this paper, we propose a new objective evaluation method which agrees with the subjective result. Moreover, we propose to adjust the coefficient adaptively in real-time according to the existence of voice in the input signal. The effectiveness of the van able coefficient is examined through computer simulations using speech signals.