Perceptual adaptation to foreign accented speech.

Abstract

Previous research on native listener adaptation to foreign-accented adaptation has demonstrated that training on a single non-native speaker results in talker-dependent learning (i.e., training does not generalize to new talkers). However, training on multiple foreign-accented speakers from a single language background results in talker-independent, but accent-dependent learning (i.e., training generalizes to new speakers of the trained accent, but not to new accents). The current study extends these findings, examining whether training on multiple accents can result in accent-independent learning. Native English listeners were trained on recognition of foreign-accented speech by five speakers from different language backgrounds. They were then tested on the speech of two novel speakers. One of the speakers was a native speaker of a language included in the training set (Mandarin), and the other was a native speaker of a language not included in the training set (Slovakian). Listeners demonstrated better performance on both the Mandarin and Slovakian tests than untrained control subjects, demonstrating accent-independent learning after training on multiple foreign accents. This suggests that individuals in multilingual communities, who are exposed to significant variation in the input, may be able to achieve highly flexible speech perception systems.