Designing a spoken corpus of CV syllables based on Persian prosody for using in TTS software package

<p dir="ltr">The most important feature of a text-to-speech program is the naturalness of the generated speech. In text-to-speech programs, the processing of sound typically involves changing the fundamental frequency and modifying the units of synthesis, resulting in speech that possesses certain acoustic characteristics. However, considering recent perceptual tests and studies conducted in the field of speech unit synthesis, it seems better to synthesize units from different and sensitive acoustic contexts and then combine them to generate speech. In this case, the resulting speech will be more natural. Therefore, the preparation of synthesis units, meaning syllables, was considered, and the written structure of the Persian CV syllables was prepared accordingly. Since addressing all the differences in the acoustic properties of syllables and accounting for all these differences in the speech synthesis program is complex and challenging, the syllables in question were derived only from four acoustic structures: 1) syllables without a vowel; 2) syllables with a vowel, with a low pitch and a non-nasal tone; 3) syllables with a vowel, with a low pitch and a nasal tone, with falling boundary tones; 4) syllables with a vowel, with a low pitch and a nasal tone, with rising boundary tones. In the Persian language, there are 23 consonants and 6 vowels, which, when combined, yield 138 (138 = 6 * 23) CV syllables. Considering the four acoustic structures, the total number of acoustic syllables is 552 (552 = 4 * 138). The production of syllables occurs continuously within sentences. The recording and preparation of the speech corpus of simple and compound syllables were based on acoustic signals and spectral analysis, taking into account specific criteria. Subsequently, the resulting syllables were stored in various files based on their acoustic positions. These data can be utilized in the text-to-speech program.</p>