Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

The COST278 Broadcast News Segmentation and Speaker Clustering Evaluation - Overview, Methodology, Systems, Results

Janez Zibert (1), France Mihelic (1), Jean-Pierre Martens (2), Hugo Meinedo (3), Joao Neto (3), Laura Docio (4), Carmen Garcia Mateo (4), Petr David (5), Jindrich Zdansky (5), Matus Pleva (6), Anton Cizmar (6), Andrej Zgank (7), Zdravko Kacic (7), Csaba Teleki (8), Klara Vicsi (8)

(1) University of Ljubljana, Slovenia; (2) Ghent University, Belgium; (3) INESC-ID/IST, Lisbon, Portugal; (4) Universidade de Vigo, Spain; (5) Technical University of Liberec, Czech Republic; (6) Technical University of Kosice, Slovak Republic; (7) University of Maribor, Slovenia; (8) Budapest University of Technology & Economics, Hungary

This paper describes a large scale experiment in which eight research institutions have tested their audio partitioning and labeling algorithms on the same data, a multi-lingual database of news broadcasts, using the same evaluation tools and protocols. The experiments have provide more insight in the cross-lingual robustness of the methods and they have demonstrated that by further collaborating in the domains of speaker change detection and speaker clustering it should be possible to achieve further technological progress in the near future.

Full Paper

Bibliographic reference.  Zibert, Janez / Mihelic, France / Martens, Jean-Pierre / Meinedo, Hugo / Neto, Joao / Docio, Laura / Mateo, Carmen Garcia / David, Petr / Zdansky, Jindrich / Pleva, Matus / Cizmar, Anton / Zgank, Andrej / Kacic, Zdravko / Teleki, Csaba / Vicsi, Klara (2005): "The COST278 broadcast news segmentation and speaker clustering evaluation - overview, methodology, systems, results", In INTERSPEECH-2005, 629-632.