Speech data providers and recorded audio
Compare proprietary audio with available open corpora only after checking the particular release, licence and task fit. Useful differentiation can come from domain coverage, acoustic conditions or conversational structure—not simply more hours.
Where proprietary audio can differ
- Low-resource languages and regional dialects.
- Domain speech: clinical dictation, call-centre dialogue, field radio.
- Real acoustic conditions (vehicles, factories) rather than studio reads.
- Multi-speaker conversation with turn-level timestamps.
Consent is the gating item
Recorded calls may contain personal data, confidential material and restricted uses. A recording permission does not necessarily cover model training. Assess the applicable lawful basis, notices, contracts and any biometric requirements; voiceprints and ordinary audio are not interchangeable legal categories. Document permissions or other applicable grounds before licensing.
Metadata to prepare
- Hours, speaker count, sample rate, channels
- Language / locale tags
- Transcript method and measured accuracy on a sample
- Consent form version per speaker
Have a dataset worth discussing?
Tell us what you own, which rights you can license and the commercial scope you have in mind.