This paper presents the investigation results in the field of time scaling of speech signal. Firstly, the perceptual evidences for speeded up and slowed down speech were analysed, both in time and spectral domain. Based on this results a new algorithm for time-scaling of speech was proposed. The algorithm is not pitch synhronous and, because of that, its application is convinient for multitalker environment. Also, flexible choise of processing parameters allows application of proposed algorithm for non speech signals. The algorithm can be easily implemented in real time using a single chip digital signal processor.