Nonlinear Audio-Integrated Active Noise Control Based on Psychoacoustic Gated Convolutional Recurrent Network

The audio-integrated active noise control (AIANC) system is a special type of active noise control (ANC) system that aims to suppress noise while broadcasting audio. Although the nonlinear effect in ANC systems has been studied for many years, current research on AIANC systems still focuses on linear approaches, which cannot match the practical systems well and will lead to performance degradation. This paper proposes a new nonlinear AIANC method based on the psychoacoustic gated convolutional recurrent network (PGCRN). First, a new AIANC system model is developed based on the gated convolutional recurrent network (GCRN). Then a psychoacoustic model is introduced to the loss function of the GCRN to improve the audio quality. Finally, a new automatic gain control (AGC) method based on the MFCC cosine similarity and the XE-NLMS algorithm is proposed for the new AIANC system to improve audio quality under low signal-to-noise rate (SNR) conditions. Experimental results show that the proposed AIANC method outperforms several conventional methods in different noisy and nonlinear environments.