Phoneme based Bangla text to speech conversion

This paper presents a phoneme based Bangla Text to Speech (TTS) Synthesis framework which uses a new approach for recording Bangla phoneme to improve the speech quality. Main objective of our method is to produce more natural speech sound during Bangla text to speech conversion. In this approach, the size of the dictionary remains small, but produces more smooth and natural sounds than any other phoneme, syllable or diphone based approaches. In the proposed framework, the voice sound of each Bangla alphabet is recorded. Afterwards, the recorded voice sound is separated into their constituent phonemes by a voice cutter. This dataset can be used for future conversion of input text into its corresponding natural sounding speech from a sequence of phonemes. Before generating the phonemes sequence for each word, we normalized the text by considering numbers, abbreviations, acronyms, currency, dates and URLs. Our implemented system supports UNICODE input for Bangla text.