Foreign language talker identification training does not generalize to new talkers

Abstract

Listeners identify talkers less accurately in a foreign language than their native language, but it remains unclear whether this is due to lack of experience identifying foreign-language talkers or whether linguistic processing reveals additional talker-specific information in speech. Here, we investigated whether two types of training improved the ability to learn to identify foreign-language talkers. Participants completed four days of talker identification training in Mandarin, an unfamiliar foreign language. Participants were assigned to either the “same-voices” condition, in which they trained on the same five voices during days 1–3, or the “different-voices” condition, in which they learned a new set of five voices on each day. Both groups learned five new voices on day 4. Talker identification accuracy improved across days 1–3 for the same-voices condition, but not for those in the different-voices condition. However, talker identification accuracy on day 4 did not differ from the day 1 baseline for either group. These results suggest that knowledge about foreign-language talkers is limited to the training set, and that, without specific linguistic knowledge, training on foreign-language talkers does not generalize to improved ability to learn to identify new foreign-language talkers.