The present experiment was designed to investigate the perceptual learning of talker's voice. Listeners were trained over a 9-day period to become familiar with a set of ten talkers (five male and five female). During each day of training, listeners were asked to identify talkers' voices from isolated words and to associate each voice with a common name. On the tenth day of testing, listeners were given a generalization test which consisted of a set of novel words produced by the ten talkers they heard in training and by ten unfamiliar talkers (five male and five female). The results showed that over 9 days of training, talker identification performance improved. However, there were large individual differences among listeners in talker learning performance. In addition, individual differences in the identifiability of voices was observed. Male voices were identified significantly better than female voices by both male and females listeners. The results of the generalization task showed that the introduction of unfamiliar voices reduced identification performance for familiar talkers. Listeners were able to judge overall familiarity, but confusions increased in identification of familiar talkers. The results suggest that talker-, listener-, and task-specific factors influence the perceptual learning of talker's voice.