Input SNR Estimation using Binary Mask in Systems based on Computational Auditory Scene Analysis

This paper presents a new approach for estimating the signal-to-noise ratio (SNR) of mixture signal, which is based on the computational auditory scene analysis (CASA). The ideal binary mask (IBM) which is generally the computational goal of CASAbased systems is used to estimate the SNR of noisy speech signal. The proposed method is evaluated using IBM and some quasi-IBM masks. The method is simple and computationally efficient. Systematic evaluations show that the proposed method results in reasonable estimation of the input SNR level in a wide range of SNR values, and probable errors in estimating IBM do not affect so much the performance of the proposed system. Furthermore, simulation results show that the proposed system outperforms previous SNR estimation methods.