Bengali Conversational Speech Dataset
2,000 hours of spontaneous two-person conversations in Bengali, recorded in Bangladesh with 200 vetted native speakers. Dual-channel audio, time-aligned human-verified transcripts and speaker metadata, licensed for commercial ASR, TTS and voice AI training.
This repository is the dataset card only. The audio is licensed commercially and isn't hosted on Hugging Face. Free samples (audio, matching transcripts and the metadata schema) are sent on request: request Bengali samples.
Specifications
| Language | Bengali |
| Recorded in | Bangladesh |
| Speech type | Spontaneous two-person conversations |
| Volume | 2,000 hours |
| Speakers | 200 vetted native speakers |
| Audio | Dual-channel, one channel per speaker |
| Transcripts | Time-aligned, human-verified |
| Speaker metadata | Age, gender, region, recording device |
| Consent | Signed commercial-use consent from every speaker |
| License | Commercial, for training and evaluation |
| Price | $65 per hour of audio, volume discounts available |
What makes Bengali hard for speech recognition
Bengali has many aspirated consonants and nasal vowels, and Bangladeshi Bengali differs from Kolkata Bengali in pronunciation and vocabulary. Colloquial speech is far from formal written Bengali, and regional varieties like Sylheti and Chittagonian are hard to follow even for other Bengali speakers.
Built for
- Automatic speech recognition (ASR / STT)
- Text-to-speech and voice synthesis prosody
- Voice agents and IVR for Bangladesh
- Speaker diarization and turn-taking models
- Speech LLM and audio-understanding fine-tuning
Getting the data
Request samples with a company email on the Bengali dataset page and the team replies with representative audio and transcripts so you can check quality, accents and transcript conventions before you buy. Need more hours, another dialect or your own recording setup? SpeechData.ai also collects speech data to spec, or you can book a call.
About SpeechData.ai
SpeechData.ai licenses conversational speech datasets in 60 languages and dialects and runs custom speech data collection. It's part of Spirelight. Browse the full catalog.
- Downloads last month
- 32