Lipi-Ghor: 885-hour diarized Bengali speech corpus from YouTube
Original title: Lipi-Ghor: A Large-Scale, Diarized, Multi-Domain Bengali Speech Corpus and a Scalable Pipeline for Building It
Audio from 998 videos across 17 domains is paired with noisy captions plus a 30-hour human gold split; raw audio is not redistributed.
Researchers released Lipi-Ghor, a weakly supervised Bengali speech corpus of 885 hours drawn from 998 YouTube videos on 596 channels. Annotations in SSTT format combine automated speaker diarization with Bengali caption tracks and cover long-form audio that existing open Bengali ASR sets lack. Four widely used ASR models degraded on emotive, dialectal, and long-form content. Video IDs, diarization, pipeline code, and gold labels are under CC-BY-4.0; caption transcripts are research-only.
- The corpus spans 17 domains, including audiobooks, drama, sermons, news, and dialects.
- Seventeen volunteers verified a gold split over 30 hours of correction.
- The pipeline is presented as reusable for other captioned YouTube languages.
- No raw audio is redistributed.
From the original article
1 min read at huggingface.co
Bengali, spoken by over 230 million people, remains severely under-resourced in spoken language research. Existing open Bengali ASR corpora are confined to short, read or crowd-sourced speech and lack speaker diarization and long-form audio.
By Sanjid Hasan, A H M Fuad, Risalat Labib · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.