Text to Speech Synthesis for Punjabi Language.

Abstract

In recent years, the use of computers in speech synthesis and speech recognition has become an important area of study among speech and computer scientists. The primary motivations are to provide users with a friendly vocal interface with the computer and to allow people with certain handicaps (such as blindness) to use the computer. The tremendous recent developments in speech and computer technology have produced unrestricted-vocabulary speech synthesis on PCs in English and some other European languages. In India, very few institutions are working in the field of speech synthesis. Indian Institute of Technology, Kharagpur, has developed an Indian language speech synthesizer named “Shruti” for Hindi and Bengali languages. Another popular speech synthesizer available for Indian languages is “Dhvani”, which is for Hindi and Kannada languages. In speech synthesis, the accuracy of information extraction is crucial in producing high quality synthesized speech. Speech synthesis involves the algorithmic conversion of input text data to speech waveforms. Speech Synthesizers are characterized by the method used for storage, encoding and synthesis of the speech. The synthesis method is determined by the vocabulary size as all possible utterances of the language need to be modeled. It is a well-established fact that text-to- speech (TTS) synthesizers designed for use in a restricted domain always perform better than their generalpurpose counterparts. The design of such general purpose synthesizers are complicated by the fact that the sound output needs to be close to natural speech. The presented work attempts to achieve this type of speech synthesis for Punjabi language. The synthesizer uses an existing TTS named Dhvani that has been developed for Hindi and Kannada Language, as a model. Dhvani acts as a prototype of a phonetics-to-speech engine, which can serve as a back-end for speech synthesizers in many Indian languages. One just has to associate it with the corresponding language specific text-to-phonetics module. The synthesizer has been implemented and tested successfully on Fedora Core 3 platform. The results are evaluated to determine the performance of the synthesizer developed in terms of intelligibility and naturalness.