Tasks/ASR / TTS

ASR / TTS

কথন শনাক্তকরণ ও সংশ্লেষণ

Bangla speech recognition and text-to-speech corpora and models.

5 datasets·100 papers
DatasetSizeLicenseYearSource paperLink
Lipi-Ghor (bn-882-SSTT)
Long-form multi-speaker Bangla ASR and speaker-diarization corpus with aligned transcriptions and timestamps. Gated access.
882 hrsCC BY 4.02026Lipi-Ghor-882 (DL Sprint 4.0, Team Villagers)
✓ verified 2026-07-22
Bengali-Loop (ASR corpus)
Long-form Bangla ASR corpus of 792k words from 11 YouTube channels, built by subtitle extraction with human-in-the-loop transcript verification.
191 recordings / 158.6 hrsMixed2026Bengali-Loop (DL Sprint 4.0, Tabib et al.)
✓ verified 2026-08-10
Common Voice (bn)
Continuously growing crowdsourced speech corpus.
1,200+ hrs recordedCC02024Mozilla Common Voice
✓ verified 2026-07-01
OOD-Speech
Large-scale out-of-distribution Bangla ASR benchmark.
1,177 hrsCC BY-NC 4.02023OOD-Speech (2023)
✓ verified 2026-07-19
OpenSLR SLR53 (Large Bengali ASR)
Crowdsourced Bangladeshi Bangla read speech.
196k utterances / 229 hrsCC BY-SA 4.02018Google crowdsourced speech
✓ verified 2026-06-28