Enhanced Time-Frequency Masking by Using Neural Networks for Monaural Source Separation in Reverberant Room Environments

Abstract

Deep neural networks (DNNs) have been used for dereverberation and denosing in the monaural source separation problem. However, the performance of current state-of-the-art methods is limited, particularly when applied in highly reverberant room environments. In this paper, we propose an enhanced time-frequency (T-F) mask to improve the separation performance. The ideal enhanced mask (IEM) consists of the dereverberation mask (DM) and the ideal ratio mask (IRM). The DM is specifically applied to eliminate the reverberations in the speech mixture and the IRM helps in denoising. The IEEE and the TIMIT corpora with real room impulse responses (RIRs) and noise from the NOISEX dataset are used to generate speech mixtures for evaluations. The proposed method outperforms the state-of-the-art methods specifically in highly reverberant and noisy room environments.