Designing and Recording a Speech Database for Persian TTS Systems

This paper presents a method for designing and building a speech database for Persian TTS systems taking account of Persian prosodic structure. This collection¡ containing 2826 phonetically and prosodically rich utterances¡ was recorded under studio conditions with a female voice talent speaker. In addition to common practice of covering phonetic units¡ the coverage of prosodic units has been considered in preparation of this collection as a novel procedure in building Persian databases. The current database consists of sound files and their corresponding manually-corrected orthographic and phonetic transcriptions. Finally¡ we used the prepared database to do speech synthesis using a statistical-parametric method. The subjects evaluated the generated voices 4.3 in terms of MOS criterion.