TTS was trained for Assamese phonology rather than adapted from a pan-Indic voice. In a blind preference study, listeners rated it clearer on alveolar/retroflex contrasts and on vowel length that machine-translated prompts often flatten. The 64.5% figure is the mean of the four prompt types below.
We report mean opinion scores and pairwise preference against a strong multilingual TTS baseline on the same Assamese prompts.