This paper reports the I2R’s submission to the Blizzard Challenge 2008. This is our first participation in Blizzard Challenge. In this paper, we describe the approach that we used to build the three required voices. We introduced the acoustic parameters that include MFCC coefficients as spectral parameters in addition to the prosodic parameters for unit selection based speech synthesis. We used regression tree approach to predict the acoustic parameters. The evaluation results show that our approach works well for English speech synthesis. The approach has also shown good performance in keeping the speaker characteristics of the speech database in Mandarin speech synthesis.