BanglaKontho: Closing the Long-Form Gap in Bangla Text-to-Speech

What changed

Researchers released BanglaKontho, a 20-hour single-speaker Bangla TTS audiobook dataset, a text normalization pipeline, and a baseline model reaching 4.46 MOS.

Why you should care

Provides long-form single-speaker audio data and normalization tools to improve under-resourced Bangla neural speech synthesis.

Your move

Test. Validate dataset suitability for long-form speech generation.

Evidence

Preliminaryarxiv.org ↗

What to watch next

Independent verification of the 4.46 MOS score across external Bengali listening tests.

Event

research

Event date

Sep 25, 2026

Relevant to

General AI readers