A Novel Intonation Model to Improve the Quality of Tamil Text-to-Speech Synthesis System

The global growth of Information and Communication technologies has a greater impact towards the research focus on speech technologies. Especially visually impaired people, vocally challenged people can utilize speech technology enabled devices as it helps them as a lifeline. In the broad sense, Speech technology has two major applications, speech synthesis and speech recognition. Speech synthesis is a popular technique to produce synthetic speech given the input text, whereas speech recognition is the technique that understands human speech and can produce either text or speech as output. Tamil Nadu is one of state in southern region of INDIA, the eleventh biggest state and it also stands seventh as a most populous state in India, has over 74 million population. Among these huge population, only 58 million people are literates, 0.3 million people are visually differently-abler. In spite of the tremendous growth in information and communication technology, there is a great demand for speech technology enabled devices in the regional language Tamil to facilitate the illiterates and especially visually challenged people to avail the technological and communicative facilities at par with others. But, speech applications in Tamil put forward a greater demand for the quality of speech. The government of Tamil Nadu encourages the researchers to innovate software's and devices that in fact help the visually challenged people to enlighten their lives. Researchers are still making efforts to produce highly intelligible and natural sounding speech irrespective of the language. This paper investigates the available prosodic models and details on the prosodic parameters that contribute towards the improvement in quality of speech synthesis systems. This study streamlines the method to develop an intonation model, which is one of the important prosodic parameter to accomplish the quality in terms of naturalness of the speech in Tamil TTS. The subjective evaluation of the proposed method shows the significant improvement in the quality in terms of naturalness of the produced speech.