Speaker change detection using joint audio-visual statistics
Giridharan Iyengar, C. Neti · 2000
In this paper, wepresentanapproach for speaker change detection in broadcast video using joint audio-visual scene change statistics. Our experiments indicate that using joint audio-visual statistics weachieve better recall without loss of precision as compared to purely audio domain approaches for speaker change detection. 1 Context In the emerging multimedia and ubiquitous computing applications there is a general awareness of the need to perform joint audio-visual analysis. Examples of such applications include speechreading #1, 6, 7, 8, 12# and general multimodal communication#3#. One of the fundamental goals in this emerging discipline is to model audio-visual events for the purposes of understanding, indexing and managing multimedia content. There is signi#cant value in generating automatic transcripts and summarization of such audio-visual content. An example of suchwould be generating a textual transcription#keywords automatically from the speech portions of the audio-visu...