A source separation approach for the joint representation of OBA and SBA under reverberation conditions

Spatial audio is implemented in three formats, i.e., Channel-Based Audio (CBA), Object-Based Audio (OBA), and Scene-Based audio (SBA). The joint representation of multiple audio formats can combine their advantages to satisfy the technical requirements of complex audio scenes. Specifically, the joint representation of object-based audio and scene-based audio plays a crucial role in spatial audio. However, the recorded SBA signal contains the associated components of OBA, which results in mutual interference during sound field reconstruction. Therefore, separating the target SBA component from the original SBA signal becomes the key to achieving the combination of the two audio formats.In fact, our previous work deals with outdoor applications, and a source separation approach for the recorded SBA signals in the joint representation of OBA and SBA under non-reverberation conditions was proposed. In this paper, a linear convolution model for the original SBA signal under reverberation conditions is firstly built up for the indoor applications. Then an NMF-based sound source separation algorithm is proposed, while a novel EM parameter initialization method is also proposed for the algorithm. Finally, the effectiveness of the proposed separation algorithm is verified by groups of experiments.