The localization of sound sources, and particularly speech, has a numerous number of applications to the industry. This has motivated a continuous effort in developing robust direction-of-arrival detection algorithms. Time difference of arrival-based methods, and particularly, generalized cross-correlation approaches have been widely investigated in acoustic signal processing. Once a probability function is obtained, indicating those directions of arrival with highest probability, the vast majority of methods have to assume a certain number of sound sources in order to process the information conveniently. In this paper, a model selection based on a Bayesian framework is proposed in order to determine, in an unsupervised way, how many sound sources are estimated together with the parameters estimation. Real measurements using two microphones are used to corroborate the proposed model.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.