Acoustic Pairing of Original and Dubbed Voices in the Context of Video Game Localization

Adrien Gresse (1) , Mickael Rouvier (1) , Richard Dufour (1) , Vincent Labatut (1) , Jean-Francois Bonastre (1)

1 LIA - Laboratoire Informatique d'Avignon

The aim of this research work is the development of an automatic voice recommendation system for assisted voice casting. In this article, we propose preliminary work on acoustic pair-ing of original and dubbed voices. The voice segments are taken from a video game released in two different languages. The paired voice segments come from different languages but belong to the same video game character. Our wish is to exploit the relationship between a set of paired segments in order to model the perceptual aspects of a given character depending on the target language. We use a state-of-the-art approach in speaker recognition (i.e. based on the paradigm i-vector/PLDA). We first evaluate pairs of i-vectors using two different acoustic spaces, one for each of the targeted languages. Secondly, we perform a transformation in order to project the source-language i-vector into the target language. The results showed that this latest approach is able to improve significantly the accuracy. Finally, we challenge the system ability to model the latent information that holds the video-game character independently of the speaker, the linguistic content and the language .

Mots clés: Speaker recognition; Voice casting; Voice similarity; i-vector; Video game

Domaines: Son [cs.SD]; Informatique et langage [cs.CL]

Adrien Gresse, Mickael Rouvier, Richard Dufour, Vincent Labatut, Jean-Francois Bonastre. Acoustic Pairing of Original and Dubbed Voices in the Context of Video Game Localization. Interspeech, Aug 2017, Stockholm, Sweden. pp.2839-2843, ⟨10.21437/Interspeech.2017-1311⟩. ⟨hal-01572151⟩