Preparation of sound base for a text-to-speech synthesis system

摘要

We are giving several recommendations for the choice of parameters of the sound fragments in this report. The sound fragments are components of the sound base, used in Russian speech synthesis system by a text. It isn't the secret that quality of concatenation synthesis in many respects is defined at the stage of a speaker choice and preparation of base of speaker's voice samples. Formulated recommendations are received on the basis of the statistic analysis of big amount of various types of texts and concern both separate sound fragments and their groups. Parameters of sounds were taken with the help of the automatic linguistic processor including phonetic and prosodic transcriptors. The duration, intensity and main pitch frequency of sounds in various contexts and intonational contours were analyzed. The sound base produced according to the worked out recommendations, allows to make better intelligibility and naturalness of synthetic speech due to minimization of changes of speaker's voice samples.