Consonant duration from an Arabic speech database

A modular software text-to-speech system for Arabic language was developed previously (1987). To improve the acceptability of the system, particular efforts were made to achieve an accurate processing of the segments duration. The aim of the present study is to formalize a set of rules to control the variations of the Arabic consonants duration. The raw statistics concerning the durational data are collected from a large scale Arabic speech database by means of a management system including the specifications of the contexts and the features associated to each phoneme. Interrogation is realized by a procedure of the type “searching by key-words”. Using the statistically significant factors relative to the difference between the model and the natural data, a duration model for Arabic consonants including compensatory adjustments between several linguistic factors at the word level is proposed. The root mean square error between the durations found in the database and the durations predicted by the model proposed is equal to 14.74 ms