Gémellité et reconnaissance automatique du locuteur

Abstract : This paper presents a speaker recognition system applied to a limited corpus of twin speakers. Two experiments are carried out using the LIA speaker recognition platform AMIRAL. The aim of the first experiment is to identify a twin of a speaker among a corpus, whereas the second one is a classical speaker verification task. The results show that an automatic system is generally able to identify a twin with an acceptable performance (85 % of good identification). Furthermore, the system is still able to discriminate the target speaker from its twin. For the verification experiment, we obtain 6% F.A. for 0% F.R. and 53% F.R. for 0% F.A.. As the available amount of data is extremely limited, results have to be interpreted with caution.
Complete list of metadatas

Cited literature [10 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-00134198
Contributor : Cnrs : Umr6057 Laboratoire Parole Et Langage <>
Submitted on : Tuesday, March 6, 2007 - 10:53:49 AM
Last modification on : Saturday, June 15, 2019 - 12:24:17 PM
Long-term archiving on : Tuesday, April 6, 2010 - 10:27:49 PM

File

1743.pdf
Publisher files allowed on an open archive

Identifiers

  • HAL Id : hal-00134198, version 1

Collections

Citation

Nicolas Scheffer, Jean-François Bonastre, Alain Ghio, Bernard Teston. Gémellité et reconnaissance automatique du locuteur. Journées d'Etudes sur la Parole, 2004, Fez, France. pp.445-448. ⟨hal-00134198⟩

Share

Metrics

Record views

471

Files downloads

374