CNN-Conformer: Conformer in Channel Mapping Based Convolutional Neural Network for Stereophonic Acoustic Echo Cancellation
Tao Liu, Hongqing Liu, Yin Liu, Yi Zhou · 2023
In acoustic echo cancellation (AEC), conventional methods estimate the path of the microphone and loudspeaker to achieve clear communication. However, for stereophonic AEC (SAEC), the problem of non-uniqueness arises because of a high correlation between far-end signals. This paper proposes a convolutional neural network (CNN) with a conformer module to remove echoes and background noise. First, the generalized cross-correlation (GCC) is utilized to align the far-end signals with the echoes in the microphone signal. After that, the complex spectral features are fed into the CNN-Conformer, where the encoder module is used to extract the frequency information, the conformer module explores the channel relationship, and the temporal convolutional network (TCN) module performs time modeling. Finally, a joint loss function in the time domain and frequency domain is designed to train the network. The experiments show that the proposed approach presents a good generalization capability and performs better than previous neural network-based approaches in SAEC.