An Integration Development of Traditional Algorithm and Neural Network for Active Noise Cancellation

Traditional active noise cancellation (ANC) methods are based on adaptive signal processing with adaptive algorithms as the foundations. They are linear systems and do not perform satisfactorily in the presence of nonlinear distortions. In this paper, we exploit the nonlinear processing capabilities of neural networks to address nonlinear ANC problems. The main idea is to employ deep learning to encode optimal control parameters corresponding to different noises and environments. We first actively remove the noise or noisy speech through the traditional adaptive filter methods, and then the residual noise and estimated noise are sent into the neural network to eliminate the nonlinear distortions. Regardless of whether the reference signal is noise or noisy speech, ANC system can be achieved by training a neural network. The experimental results show that the proposed ANC system demonstrates a good noise reduction effect on complex noises, and produces a superior performance compared with the traditional methods.