Research on Speech Signal De-Noising Algorithm Based on Variable Step Size LMS Adaptive Filter with Strong Background Noises
Guangyan WANG,Zhenhao YANG,Lei CHEN,Haifeng LI
DOI: https://doi.org/10.1117/12.2631958
2022-01-01
Abstract:Aiming at the relatively difficult problems of speech signal enhancement and de-noising under low signal-to-noise ratio and strong background noise, this paper adopts three variable step size LMS adaptive filtering algorithms: variable step size LMS algorithm based on sigmoid function product, variable step size LMS algorithm based on adjusting the optimal exponential factor of sigmoid function and variable step size LMS algorithm based on sigmoid function feedback de-noise The speech signals in different noise environments such as factory noise and ocean noise are de-noised. After the pure speech signal is superimposed with - 10dB, - 5dB, 0dB and 5dB Gaussian white noise, pink noise, factory noise and marine noise respectively, the variable step size LMS adaptive noise cancellation processing is carried out. SNR, PESQ and other objective evaluation algorithms are used to evaluate the sound quality of the enhanced speech signal. The simulation and calculation results show that the three variable step size LMS algorithms used in this paper have good de-noising effect on speech signals with strong background noises of low signal-to-noise ratio, and the variable step size LMS algorithm based on sigmoid function feedback control has better de-noising effect than the other two algorithms under this condition.