Multi-source DOA estimation in a reverberant room

In a reverberant scenario, phase transformed weighted algorithms are more robust than maximum likelihood (ML) because of the insufficiency of the data model to incorporate reverberant information. This transformation has been applied to general cross-correlation and steered response power algorithms; the latter has been shown to be more robust. For a multiple known number of sources, both algorithms have problems separating the sources that are close together because of the limitation caused by the resolution. Recently, an approach was made using simple well-known statistical room acoustics to model room reverberation, and another parametric approach called the approximate maximum likelihood was made that was designed for multi-source estimations. By combining these methods, we developed an ML algorithm that is suitable for multi-source target estimates in a reverberant room.