Determining the precise circumstances under which non-linguistic information such as talker identity influences spoken word recognition is critical for models that postulate both abstract and detailed representations. In this study, we used the long-term repetition priming paradigm and manipulated both talker identity and word frequency. The results showed that change in the talker identity diminishes the magnitude of the repetition priming effect when words are of low frequency but not when they are of high frequency. Our results support the time-course hypothesis (McLennan & Luce, 2005) which posits that talker specificity effects take time to develop, and thus are more likely to emerge when lexical identification requires time to be accomplished.