Maithili Text-to-Speech System

This paper discusses development of TTS system for Maithili language. Speech corpus spanning 5 hours borrowed from LDCIL, CIIL Mysore, and also of 3 hours is collected from native speakers in studio environment. As most Indian languages including Maithili are syllabic in nature, concatenative method is used for the purpose of speech generation taking syllable as a basic unit. To enhance naturalness of speech out, 1055 most frequently occurring words have been recorded and stored. The system supports UTF-16 for text input. C#.NET is used for development of interface. The speech database consists of 930 syllable (C*V) in total. Each position has 300 syllables and 10 independent vowels. 930 units of speech data is built from all three positions. Subjective Evaluation, MOS and MRT, are conducted by 10 native speakers. The quality of synthesized speech in terms of intelligibility and naturalness is evaluated to be approximately 84 percent. The relevance of the work lies in the fact that no TTS system exists for Maithili Language till date.