Bilingual talker identification with spontaneous speech in Cantonese and English: The role of language-specific knowledge

Abstract

Previous research has shown that implicit or explicit knowledge about a language lends itself to improved talker recognition performance. In the context of this language familiarity effect, it is apparent that bilinguals are better at generalizing voice learning across their known languages compared to monolingual listeners [Orena et al., JASA, 146 (2019)]. Other work suggests that training in an unknown language generalizes to a known language more robustly than the reverse [Winters et al., JASA 123 (2008)]. The current study launches from these previous studies. This project uses excerpted Cantonese and English snippets from spontaneous interview speech from Cantonese-English bilinguals in a talker identification training experiment with Cantonese-English bilingual listeners and bilingual listeners with no knowledge of Cantonese (or related languages). Listeners are assigned to either Cantonese or English training, and then all listeners are tested on both Cantonese and English utterances to assess learning for the trained language and generalization to the bilingual's second language. Results of a multilingual questionnaire quantify listeners' code-switching abilities and multilingual competence, which, given prior research, should account for some individual differences. Using spontaneous productions, as opposed to read speech, improves the ecological validity of this research and broadens its implications.