Lipi-Ghor Dataset Releases 885 Hours of Diarized Bengali Speech
October 25, 2026
Lipi-Ghor is a new weakly-supervised Bengali speech corpus consisting of 885 hours of audio from 998 YouTube videos. It features speaker diarization and timestamped transcriptions across 17 domains, including news and regional dialects, addressing the lack of long-form Bengali ASR data.
HOW THIS AFFECTS YOU
●
researcherYou can now train more robust Bengali ASR systems using large-scale, diarized, long-form audio.