Speech datawith provenance.
Commission speech datasets where every clip traces back to a consenting person and a licence you can show your legal team.
The speech we’re building toward.
Read prompts with exact transcripts.
Please set a reminder for the market on Thursday morning.
Exact transcript, read at a natural pace
Every dataset ships with its paperwork.
- Consent record
- Per speaker: what they agreed to, when, and under which licence version.
- Licence
- Clear commercial terms for training, evaluation or both, with a traceable chain of rights.
- Transcripts
- Verbatim or normalised to your convention, with human quality review.
- Metadata
- Language, locale, self-described speaker attributes, device and recording environment.
- Audio
- Lossless files at the sample rate and channel layout your pipeline expects.
{
"clip_id": "ex_01HZX…",
"language": "yo",
"locale": "yo-NG",
"task": "scripted",
"transcript": "…",
"duration_s": 6.42,
"audio": { "format": "wav", "sample_rate": 48000, "channels": 1 },
"speaker": {
"id": "spk_…", // pseudonymous
"age_band": "25-34", // self-described
"accent": "Lagos" // self-described
},
"environment": "indoor_quiet",
"device": "smartphone",
"consent": {
"record_id": "cns_…",
"licence": "echo-train-eval-v1",
"granted_at": "…"
},
"qa": { "reviewed": true, "transcript_match": "verified" }
}How a commission works
Scope
We agree languages, locales, speaker mix, recording conditions, annotation and licence terms with your team.
Collect
Contributors who match the specification record through the ECHO app, with consent captured per task.
Verify
Every clip passes automated checks and human review for audio quality, transcript accuracy and specification fit.
Deliver
You receive audio, transcripts, metadata and consent records in the format your pipeline expects.
Tell us what you need.
Share as much or as little as you know. We’ll come back with questions, feasibility and next steps.