On-Line Multi-Modal Speaker Diarization

This paper presents a novel framework that utilizes multimodal information to achieve speaker diarization. We use dynamic Bayesian networks to achieve on-line results. We progress front a simple observation model to a complex multimodal one, as more data becomes available. We present an efficient way to guide the learning procedure of the complex model using the early results achieved with the simple model. We present the results achieved in various real-world situations, including videos coming front webcameras, human computer interaction and video conferences.