Computationally Effic Modification of Speech Us

Among the conventional time-scale modification methods [1][6], the synchronized overlap and add (SOLA) method [4] is used widely because of its good performance with relatively low computational complexity. But the SOLA method still requires much computation in evaluating the normalized crosscorrelation function for synchronization procedure [9]. In this paper, we employ 3 level center clipping method in order to reduce the computational complexity of SOLA method. The result of subjective preference test indicates that the proposed method can reduce computational complexity by over 80% comparing with the conventional SOLA method without considerable performance degradation. We also apply the variable time-scale modification method using transient information [7] to the proposed algorithm. By doing so, we can maintain the intelligibility of time-scale modified speech in the case of very fast playback.