A text-to-speech system for Arabic using neural networks

Text-to-speech synthesis is of great interest and its applications are several. For this reason, it has interested many researchers for decades. Two methods are usually used: synthesis by rule and synthesis by concatenation of pre-recorded sounds. But these methods have some disadvantages such as difficulty to be adapted to a new speaker or to a new language. Recently, neural networks (NN) have been used with nonconventional problems where a traditional solution seems impossible. Text-to-speech appears as one of these problems. In this field, it has been shown that NN don´t work well when they are directly fed with speech samples. Therefore, works have been done to explore and evaluate different parametric forms of speech based on linear predictive coding (LPC), used for training, and found that LSP produced the best results. However, these methods don´t take into account residual signal and speech produced was machine-like and not natural. We propose in this paper to drive the NN with codebook-excited linear prediction (CELP), which provides high quality speech, to perform Arabic speech synthesis