Bani TTS was trained for Assamese phonology rather than adapted from a pan-Indic voice. Listeners in a blind preference study consistently rated it clearer on alveolar/retroflex contrasts and on vowel length that machine-translated prompts often flatten.
We report mean opinion scores and pairwise preference against a strong multilingual TTS baseline on the same Assamese prompts.
Deep model documentation stays on the Models page. This post is the release record.