Skip to Main content Skip to Navigation

Structuration automatique de talk shows télévisés

Abstract : Archives professionals have high expectations for efficient indexing tools. In particular, the purpose of archiving TV broadcasts has created an expanding need for automatic content structuring methods. In this thesis, is addressed the task of structuring a particular type of TV content that has been scarcely studied in previous works, namely talk show programs. The object of this work is examined in the light of a number of sociological studies, with the aim to identify relevant prior knowledge on the basis of which the structuring approach is motivated. Then, having highlighted that a structuring scheme should be assessed according to specific use cases, a user-based evaluation is undertaken. The latter stresses out the relevance of considering the speakers' interventions as elementary structural units instead of video shots usually employed in similar studies. Having emphasised the importance of speaker oriented detectors, the second part of this thesis is thus put on speaker diarization methods. We first propose a state of the art of the techniques -- particularly unsupervised ones -- used in this research domain. Then, results on a first speaker diarization system are presented. Finally, a more original system exploiting efficiently audiovisual information is finally proposed. Its validity is tested on two talk show collections : Le Grand Échiquier and On n'a pas tout dit. The results show that this new system outperforms state of the art methods. Besides, it strengthens the interest of using visual cues--even for tasks that are considered to be exclusively audio such as speaker diarization--and kernel methods in amultimodal context.
Complete list of metadata

Cited literature [172 references]  Display  Hide  Download
Contributor : Félicien Vallet Connect in order to contact the contributor
Submitted on : Tuesday, October 25, 2011 - 2:18:12 PM
Last modification on : Friday, July 31, 2020 - 10:44:06 AM
Long-term archiving on: : Sunday, December 4, 2016 - 4:52:22 AM


  • HAL Id : pastel-00635495, version 1



Vallet Félicien. Structuration automatique de talk shows télévisés. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2011. Français. ⟨pastel-00635495⟩



Record views


Files downloads