Abstract: By analyzing the present mainly prefetch algorithms advantages and disadvantages, ccording to the difference between VoiceXML-based voice platform and HTML-based World Wide Web, it is proposed that the voice resource should be prefetched in VoiceXML-based voice platform. An adaptive multi-user shared predict Markov model is presented to predict the probability of the forthcoming required resource of all the online users, which help to improve the veracity of the prediction. Finally, a preemptive priority model is designed to schedule the prefetch tasks, which mapped the resource access probability to the task priority. The simulation research shows that the predict algorithm and the schedule model can reduce delay of a users request and improve response speed better than that of single-user predict Markov model and Round-Robin (RR) schedule model.