Design of the Wiener gain in noisy and reverberant environments

The Wiener gain or filter plays an important role in both single-channel and multichannel speech enhancement. Traditional Wiener gain formulations are typically based solely on either the signal-to-noise ratio (SNR) or the coherent-to-diffuse ratio (CDR), rendering them suboptimal in noisy and reverberant environments. In this paper, we present an approach to the design of the Wiener gain by incorporating estimates of both SNR and CDR. Additionally, we introduce two hyperparameters to govern the extent of noise reduction and reverberation suppression. Through simulations, we demonstrate that the proposed approach surpasses widely used methods in terms of SNR gain, log-spectral-distortion (LSD), direct-to-reverberant energy ratio (DRR), and Kurtosis ratio (KR).