Subword based approach for grapheme-to-phoneme conversion in Bengali text-to-speech synthesis system

In this paper, we propose a subword based approach for grapheme-to-phoneme (G2P) conversion in a text-to-speech (TTS) synthesis system. The proposed method resolves the problems present in both the manual and rule-based approaches for G2P conversion. The subword method uses a segmentation procedure which chops a word into its main part (root word) and subword part (suffix). By proper segmentation of a word, this method efficiently offers insight into the basic morphological information of the word. The proposed method reduces the size of the pronunciation dictionary without affecting the vocabulary coverage. For the rule-based approach, the proposed method segments the word into two parts and predicts the pronunciation for both the word segments separately. The final pronunciation is achieved by concatenating pronunciations for both the word segments. The subword method improves the accuracy of the rule-based approach by resolving the ambiguity especially in case of the inflected or compound words.