Blind 2-microphone acoustic noise reduction algorithms using efficient variable step-size adapted by minimizing the intercorrelation function

Recent advancements in adaptive noise signal reduction have utilized 2-microphones adaptive algorithms. Specifically, the normalized form of least-mean-square algorithm (NLMS) with fixed-step-size parameters (FS) has been combined with direct-and-recursive structures of source separation. Compared to conventional one-microphone methods, these combinations provide superior speech quality. However, the main limitation of these 2-microphones adapting algorithms (Direct combination: Forward NLMS and Recursive combination: Backward NLMS) lies in their poor steady state regime with large FS value, while small step-sizes values result a slow speed of convergence. To address these issues, we propose a new variable step-size (VS) approach in this study, based on minimizing the intercorrelation function in the time domain for the basic FNLMS and BNLMS algorithms. Our approach is proposed exactly to determine an optimal value of VS parameters by minimizing the intercorrelation between the enhanced signal and the noisy microphone signals. These methods improve steady state values and convergence speed at the same time. The proposed 2-microphones adapting algorithms were evaluated through simulations conducted in high-noise environments, using the system of mismatch criterion and estimation of output segmental signal-to-noise ratio ones. The comparative simulations results confirmed that our algorithms outperform FS algorithms in terms of steady state values and convergence speed.