Statistical parametric speech synthesis for Arabic language using ANN

Statistical parametric approach for speech synthesis becomes more popular over the concatenative approach due to the low size of the system and the high-quality speech. Moreover, few researches have been done in the field of speech synthesis for Arabic language with a poor quality of speech. In this paper, we propose a statistical parametric synthesis system for Arabic based on Artificial Neural Networks (ANN). Mel frequency Cepstral coefficients (MFCC), F0, energy and duration are the main components of our system. Speech waveform is generated from the predicted parameters F0, energy and MFCC. Different methods are proposed for this development process. In addition, we propose a method to solve the problem of discontinuities between neighboring segment boundaries in order to improve the speech quality. Experimental results of cepstral and prosodic parameters are given in this paper as well as the subjective evaluation.