19 th INTERNATIONAL CONGRESS ON ACOUSTICS MADRID, 2-7 SEPTEMBER 2007 A CROSS-LINGUISTIC PERCEPTUAL AND ACOUSTICAL STUDY OF SPEAKER DISCRIMINATION IN FILM DUBBING PACS: 43.71.Hw Payri, Blas 1 ; Enríquez Carrasco, Emilia V. 2 ; Redondo, Javier 3 ; Picó, Rubén 4 Universidad Politécnica de Valencia, Campus de Gandía, 46730 Grao de Gandía, Spain; 1 bpayri@har.upv.es; 2 evenriq@idm.upv.es ; 3 fredondo@fis.upv.es; 4 rpico@fis.upv.es ABSTRACT Our experiment explores the relative importance of acoustic and linguistic features in the ability to recognize whether two utterances come from the same or different speakers, in the listener’s native language versus a foreign language. We used a set of 95 native Spanish speakers with a high-school level of English as a foreign language. The stimuli were pairs of short speech excerpts from male speakers coming from English-speaking movies, and the corresponding speech excerpts from the Spanish-dubbed movies. Finally, some excerpts from Spanish- speaking movies were used. Subjects had to rate from 1 to 6 if the two speech samples of a pair corresponded to the same speaker. The results show that listeners perform much better in their native language, with a correlation between perceptual answers and reality that is significant to the level of 0,99, whereas in the foreign language, listeners did not produce answers well correlated to reality. We discuss that in natural spontaneous speech listeners rely on linguistic features that are lost in a foreign language, even though our measures show that the acoustical diversity is similar in both versions. INTRODUCTION The experience of this paper is aimed at better understanding the issue of speaker discrimination in a foreign versus a native language. Some research tends to show that some perceptual parameters can be language-independent: for instance [11] compared the answers for Italian and German listeners using utterances in Italian and German; the results show that there is no significant difference in the ratings of age between for native and foreign language listening. EXPERIMENTAL SETTINGS Sound material To build the sound material of our experiment we have used movies with an original version in English ([1][2][3][4][5][6]), the dubbed version in Spanish of the same movies and completed it with movies with an original version in Spanish ([7][8]). All the speakers that have been chosen are male, as speaker neutralization is a phenomenon more likely to occur with same gender speakers. For each of the actors excerpts were chosen in English. Once the English-version excerpt chosen, the corresponding Spanish-version excerpt was chosen, ensuring that the same spoken content was respected. 50 English-version and the corresponding 50 Spanish- version excerpts were thus chosen. To complete the sound material, excerpts from movies that had been originally shot in Spanish for a total of 18 Spanish original version excerpts. The total amount added up to 68 excerpts in Spanish and 50 excerpts in English. Regarding the speakers, 17 actors were used in English and 24 actors in Spanish including the dubbing actors and the original actors of Spanish movies. The excerpts were chosen with the following conditions: • The speech material must form a sentence. In particular, longer sentences were not sliced to be shortened if that provoked a disruption in the prosody resulting in either the feeling of an “unended” or “unstarted” sentence. Most sentences have a verb and all are grammatical both in English and in Spanish.