Conjugate Mixture Models for Clustering Multimodal Data