Acoustic Echo Cancellation with the Normalized Sign-Error Least Mean Squares Algorithm and Deep Residual Echo Suppression

This paper presents an echo suppression system that combines a linear acoustic canceller (AEC) with deep complex convolutional recurrent network (DCCRN) for residual suppression. The filter taps of the AEC are adjusted in subbands by using normalized sign-error least mean squares (NSLMS) algorithm. NSLMS is compared commonly-used (NLMS), and combination each proposed model studied. utilization pre-trained deep-learning speech denoising as alternative to suppressor (RES) also results showed performance superior NLMS all settings. With output, RES achieved better than larger denoiser model. More notably, performed considerably on output gap was greater respective when employing RES, indicating more akin noise speech. Therefore, little data available train viable preceding AEC.