Towards a better integration of written names for unsupervised speakers identification in videos

Abstract : Existing methods for unsupervised identification of speakers in TV broadcast usually rely on the output of a speaker diariza- tion module and try to name each cluster using names provided by another source of information: we call it "late naming". Hence, written names extracted from title blocks tend to lead to high precision identification, although they cannot correct er- rors made during the clustering step. In this paper, we extend our previous "late naming" ap- proach in two ways: "integrated naming" and "early naming". While "late naming" relies on a speaker diarization module op- timized for speaker diarization, "integrated naming" jointly op- timize speaker diarization and name propagation in terms of identification errors. "Early naming" modifies the speaker di- arization module by adding constraints preventing two clusters with different written names to be merged together. While "integrated naming" yields similar identification per- formance as "late naming" (with better precision), "early nam- ing" improves over this baseline both in terms of identification error rate and stability of the clustering stopping criterion.
Document type :
Conference papers
Liste complète des métadonnées

Cited literature [14 references]  Display  Hide  Download

https://hal.inria.fr/hal-00953089
Contributor : Marie-Christine Fauvet <>
Submitted on : Monday, March 3, 2014 - 4:29:47 PM
Last modification on : Friday, March 22, 2019 - 12:56:02 PM
Document(s) archivé(s) le : Saturday, May 31, 2014 - 10:45:49 AM

File

POIGNANT--SLAM--2013.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00953089, version 1

Citation

Johann Poignant, Hervé Bredin, Laurent Besacier, Georges Quénot, Claude Barras. Towards a better integration of written names for unsupervised speakers identification in videos. First Workshop on Speech, Language and Audio in Multimedia, SLAM, 2013, Marseille, France. ⟨hal-00953089⟩

Share

Metrics

Record views

373

Files downloads

153