Abstract: The conventional SOLA method of time-scale modification encounters the problem that the higher the modification rate,the less intelligible the time-scale modified speech signal,because of the negl-ect of different contributions to articulation of different speech signal parts. This paper proposes a parti-tion time-scale modification method based on the knowledge that how fast spectrum changes and how much energy the signal contains,and both play a critical role in speech perception. After identifying portions with different spectrum and energy of a speech signal,the proposed method applies timescale modification to different portions with different modification rate. The result of subjective preference test indicates that the performance of the proposed method is superior to that of the conventional SOLA method.