Text-To-Speech (TTS) Conversion for Gujarati Language

Abstract

Text-To-Speech (TTS) conversion is a great topic of research nowadays. This project will be very useful for the illiterate and especially for the blind people. By using this system any person can read any article and understand it. The overall process of Text-To-Speech (TTS) conversion can be divided into mainly three blocks: Text Normalization, Text-To-Phoneme and Phoneme-to-Sound. Text normalization block removes the symbols and replace it with blank space and analyze all the digits and texts from the input texts. Linguistic analysis block provides intonation and prosody to the graphemes. And finally, waveform generation block will generate the original output sound. Text-To-Speech (TTS) conversion can be classified in three ways mainly: Concatenate synthesis, Formant synthesis, Hidden Markov Model (HMM) and Articulatory synthesis. As the name suggest, concatenative synthesis concatenates the different words form the database of pre-recorded words and then maps each of the matched input words with its equivalent phoneme. Due to large memory requirement of this system, this synthesis technology mainly not used in embedded systems, where memory and power is the main factors of the whole system. Formant synthesis works very well without any kind of pre-recorded word’s database. But, the drawback of this synthesis technology is that the naturalness in the output sound (robotic or not like human). HMM based synthesis is a synthesis method based on hidden markov models, also called Statistical Parametric Synthsis. In this system, the frequency spectrum (Vocal tract), fundamental frequency (vocal source) and duration (prosody) of speech are modeled simultaneously by HMMS. Speech waveforms are generated from HMMs themselves based on the maximum likelihood criterion. Articulatory synthesis is an ideal synthesis technique, in which, the whole articulatory system (mouth) of human being is modeled and sound will generate by the program. Programming point of view, this technique is very complex as compared to above methods and output is not that much accurate. Due to its complexity, this technique is rarely used for TTS system. Thus, by considering all the feasible parameters like simplicity, complexity, power, memory etc., concatenative synthesis is superior than other ones.