An Efficient Network for Farsi Text to Speech Conversion Using Vowel State
Ehsan Rasekh, Mohammad Eshghi · 2006
The main problem in Farsi text to speech synthesizers is unwritten short vowels in Farsi orthography. In this paper an ANN is used to determine the phonemes in a Farsi text. The output of this ANN is a new variable called vowel state, instead of a phoneme. Five vowel states are enough to extract pronunciations in a Farsi text, where the number of phonemes is about 30. This reduction of the output causes the reduction of interconnections of the network, considerably. The proposed vowel states approach and the ANN is tested over 2024 words with different percentage of the database as the training set. The 80.31% to 97.34% correct results are achieved using this system