Align-ULCNet: Towards Low-Complexity and Robust Acoustic Echo and Noise Reduction

The successful deployment of deep learning-based acoustic echo and noise reduction (AENR) methods in consumer devices has intensified interest in low-complexity solutions that achieve robust performance in real-world scenarios. In this work, we propose a Kalman filter - deep neural network hybrid method for AENR, which employs a novel channel-wise sampling-based feature reorientation method and time alignment in the latent space for complexity reduction and robust performance. Experimental results show that the proposed method achieves better echo reduction and comparable noise reduction performance to the considered baseline methods with improved generalization capability in many acoustically adverse scenarios, such as nonlinear distortions, low-pass filtering, and large acoustical delays.