Sylheti Nagri Text-to-Speech Synthesis

This research introduces a novel text-to-speech (TTS) system for the endangered Sylheti Nagri language, spoken primarily in the Sylhet region of Bangladesh. The Sylheti Nagri script, dating back to the 1500s, is largely obsolete today, replaced by the Bangla script. To aid in its preservation, we present the Sylheti Nagri TTS Corpus, a dataset containing over 15 hours of audio recordings, including 8,268 sentences spoken by a professional voice artist. Using the Variational Inference Text-to-Speech (VITS) framework, our model achieves a Mean Opinion Score (MOS) of 3.76, providing a crucial resource for preserving the language's oral heritage.