Telugu is one of the oldest languages in India. This paper describes the development of Telugu Text-to-Speech System (TTS) using vowel classification. Vowels are most important class of sound in most Indian languages. The duration of vowel is longer than consonants and is most significant. Here vowels are categorized as starting middle and end according to the position of occurrence in a word. The algorithm developed by us involves analysis of a sentence in terms of words and then symbols involving combination of pure consonants and vowels. Wave files are being merged as per the requirement to generate the modified consonants influenced by deergalu (vowel sign) and yuktaksharas generate the speech from a text. Speech unit database consisting of vowels (starting, middle and end) and consonants is developed. We evaluated our TTS using Mean Opinion Score (MOS) for intelligibility and voice quality with and without using vowel classification from sixty five listeners, and got better results with vowel classification.
Chaudhur, Pamela and Kumar, K Vinod
"Vowel classification based approach for Telugu Text-to-Speech System using symbol concatenation,"
International Journal of Computer and Communication Technology: Vol. 2
, Article 5.
Available at: https://www.interscience.in/ijcct/vol2/iss1/5