DEVELOPMENT OF A PROSODIC DATABASE FOR STANDARD ARABIC

The quality of a Text-To-Speech (TTS) synthesis system depends on the naturalness, and the intelligibility, of the generated speech. For this reason, a high quality automatic generator of prosody is necessary. Among the most recent developments in this field, there is a growing interest for machine learning techniques, such as neural networks, classification, and regression trees, Hidden Markov Models (HMM), and other stochastic methods. All these techniques are based on the analysis of phonetically and prosodically labeled speech corpora. The objective of our research is to realize a prosodic speech database for Arabic such as those available for other languages (the databases TIMIT for English and BDSONS for French, etc.).