An approach to recognize and pronounce words with alternative pronunciations in Farsi

摘要

In Farsi orthography some words have more than one pronunciation which corresponds to different meanings. For a good text to speech system, the words with alternative pronunciation should be determined. The proposed system in this paper is capable of recognizing and pronouncing the words with alternative pronunciations. A new definition of parameter Vowel State (VS) is used to determine the phonemes of a word. A multi layer perceptron neural network with 48, 150 and 7 neurons in the input layer, the hidden layer and the output layer is chosen to extract the phonemes. Comparing with other reported works which employ neural networks the proposed network shows efficient results according to the number of interconnections and performance. The proposed network is tested over 2024 words and results show a performance index of 85% to 95% depending on the percentage of the training set.