Implementation and Assessment of Joint Source Separation and Dereverberation

Reverberation is known to introduce difficulties in audio source separation, and reverse engineering independent sources from a convolutive mixture is one of the toughest challenges within blind source separation. This paper proposes two novel methods that combine dereverberation work with microphone interference reduction. The results are evaluated objectively using the BSS Eval toolbox and Reverb Workshop Evaluation Toolbox, relative to the effectiveness of the dereverberation and source separation. Both proposed methods show improvements on the existing dereverberation technique used. However, this has a negative impact on the source separation, as has also been seen in other work. An explanation for this negative impact and alternative approaches to avoid this situation are proposed.