Syllable specific target cost formulation for syllable based text-to-speech synthesis in Bengali

This paper proposes syllable specific target cost formulation for syllable based text-to-speech synthesis in Bengali language. Target cost formulation is implemented in three stages. In the first stage, syllable classification is done based on the position of syllable in word as begin, middle and end. At the second stage, begin, middle and end syllables are separately clustered for each of the unique syllables. Here clustering is performed with Classification and regression trees (CART) using syllable specific positional, contextual and phonological features. Target cost computation is carried out in the third stage by defining appropriate sub costs based on the above mentioned syllable specific features. All three stages in the target cost formulation prompt the unit selection module to select the best units from the database. Subjective listening tests performed on the synthesized speech shows a clear improvement in the overall quality of synthesis.