Convolutive blind source separation with low latency
Jiawen Chua, Ganlong Wang, Willem Bastiaan Kleijn · 2016
In this paper we present a convolutive blind source separation (BSS) method using crossband filtering in the short-time Fourier transform (STFT) domain. It overcomes a major limitation of the time-frequency (TF) domain BSS algorithms, which is that the window length for the STFT must be at least twice the length of the room impulse response. This makes it more appropriate to real-time applications, which require short system latency. Our experimental results show that the new method can be integrated with any TF-domain BSS algorithms and achieves better performance.