A computational modeling study was conducted using multinomial logistic regression to predict whether exposure to an unfamiliar regional accent of English would influence vowel categorization in (1) the exposure accent, (2) the native accent, and (3) another unfamiliar accent. We manipulated the number of talkers in the exposure data to determine whether talker variability influenced the efficacy of the training. Results showed a multiple-talker training benefit for the categorization of some vowels. Training also transferred to an untrained accent. Finally, the models predicted that exposure to an unfamiliar accent has a negative impact on vowel categorization in the native accent.