Low-latency real-time independent vector analysis using convolutive transfer function

Most blind source separation (BSS) methods use a long short-time Fourier transform (STFT) frame size in highly reverberant environments, which leads to a high algorithmic latency. In addition, the separated signal using the back-projection technology in most BSS methods are source images, and an extra dereverberation processing is required to obtain dereverberated sources. In this paper, we propose a low-latency real-time independent vector analysis (IVA) method using the convolutive transfer function (CTF), where the room impulse response (RIR) is approximately represented by the CTF approximation. The CTF approximation enables representations of long RIRs using a short STFT frame size, and hence the latency of the proposed method is reduced by shortening the STFT frame size in highly reverberant environments. Another advantage of the proposed method is that the dereverberated source can be estimated directly using the derived generalized back-projection technology without extra dereverberation processing. Moreover, the proposed method reduces to the original IVA in the determined case. The proposed method is compared to the state-of-the-art weighted prediction error based IVA, and the results confirm the effectiveness of the presented method in three different reverberation times.