Arabic HMM-based speech synthesis

Abstract

This paper describes the Arabic system synthesis on hidden Markov models (HTS). Our developed synthesis system uses phonemes as HMM synthesis unit, Arabic database was developed for the first test. The main objective is to maintain the consolidated text coherence which is interpreted by concatenating HMM phoneme. In our experiments, spectral properties were represented by Mel cepstrum coefficients. For the waveform synthesis, a noise or pulse excited corresponding MLSA filter was utilized. Besides that basic setup, a high-quality analysis/synthesis system STRAIGHT was employed for more sophisticated speech representation. This method has several advantages. As it is parametric, it is possible to play on the HMM parameters, change the producer voice characteristics. The developed model improves the speech synthesis, naturalness and intelligibility quality in the Arabic language environment.