Mesure de similarité fondée sur des réseaux de neurones siamois pour le doublage de voix

Abstract : Dubbing aims to broadcast a multimedia document to a larger audience. The process that consists in selecting a voice in a target language is referred as voice casting and it is performed by a human. This selection is not only based on acoustic similarity between two voices. Actually, it is suppported by more subjective criteria such as emotions, sociocultural choices... In this paper we propose a siamese neural networks based approach measuring proximity between the original voice and the dubbed one. The concept of similarity we want to model does not only consider the acoustic part of a voice, also it takes into account spectators receptive concerns. We perform a statistical test to evaluate our model. Our results show that there is an information in the acoustic parameters that allows a voice to be associated with another one with respect to a particular character.
Document type :
Conference papers
Complete list of metadatas

Cited literature [18 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01819198
Contributor : Adrien Gresse <>
Submitted on : Wednesday, June 20, 2018 - 10:06:26 AM
Last modification on : Saturday, June 15, 2019 - 12:24:17 PM
Long-term archiving on : Tuesday, September 25, 2018 - 4:35:16 AM

File

main_final.pdf
Files produced by the author(s)

Identifiers

Collections

Citation

Adrien Gresse, Richard Dufour, Vincent Labatut, Mickael Rouvier, Jean-François Bonastre. Mesure de similarité fondée sur des réseaux de neurones siamois pour le doublage de voix. XXXIIèmes Journées d’Études sur la Parole (JEP), Jun 2018, Aix-en-Provence, France. ⟨10.21437/JEP.2018-2⟩. ⟨hal-01819198⟩

Share

Metrics

Record views

153

Files downloads

187