Graduate School and Research Center in Digital Sciences

A Multimodal approach to initialisation for top-down speaker diarization of television shows

Bozonnet, Simon; Vallet, Félicien; Evans, Nicholas; Essid, Slim; Richard, Gaël; Carrive, Jean

EUSIPCO 2010, 18th European Signal Processing Conference, August 23-27, 2010, Aalborg, Denmark

This paper presents a new multimodal approach to speaker diarization of TV show data. We hypothesize that the intraspeaker variation in visual information might be less than that in the corresponding acoustic information and therefore might be better suited to the task of speaker model initialisation. This is an acknowledged weakness of the computationally efficient top-down approach to speaker diarization that is used here. Experimental results show that a recently proposed approach to purification and the new multimodal approach to initialisation together deliver 22% and 17% relative improvements in diarization performance over the baseline system on independent development and evaluation datasets respectively.

Document Bibtex

Title:A Multimodal approach to initialisation for top-down speaker diarization of television shows
Type:Conference
Language:English
City:Aalborg
Country:DENMARK
Date:
Department:Digital Security
Eurecom ref:3120
Copyright: © EURASIP. Personal use of this material is permitted. The definitive version of this paper was published in EUSIPCO 2010, 18th European Signal Processing Conference, August 23-27, 2010, Aalborg, Denmark and is available at :
Bibtex: @inproceedings{EURECOM+3120, year = {2010}, title = {{A} {M}ultimodal approach to initialisation for top-down speaker diarization of television shows}, author = {{B}ozonnet, {S}imon and {V}allet, {F}{\'e}licien and {E}vans, {N}icholas and {E}ssid, {S}lim and {R}ichard, {G}a{\"e}l and {C}arrive, {J}ean }, booktitle = {{EUSIPCO} 2010, 18th {E}uropean {S}ignal {P}rocessing {C}onference, {A}ugust 23-27, 2010, {A}alborg, {D}enmark }, address = {{A}alborg, {DENMARK}}, month = {08}, url = {http://www.eurecom.fr/publication/3120} }
See also: