Custom speech datasets recorded with consent in Pakistani homes, markets, offices and phone lines, with speaker metadata and a clear licence for AI training or evaluation.
Prompted sentences for pronunciation coverage, wake words and domain vocabulary.
Unscripted monologues on everyday topics: farming, prices, family, travel.
Natural dialogue with overlaps and turn-taking, recorded on separate channels where needed.
Speech recorded in markets, roadsides, vehicles and homes, with the noise type logged.
8 kHz and 16 kHz call-style recordings for contact-centre and voice-agent use.
Banking, health, agriculture or retail phrases written with you and read by local speakers.
We balance speakers by the factors that matter for your model and record those factors per clip.
| Factor | How we record it |
|---|---|
| Language and dialect | Self-reported and verified by a native reviewer (e.g. Multani or Derawi Saraiki) |
| District | Home district and recording location |
| Age band and gender | Broad bands only, collected with consent; never names or ID numbers |
| Recording condition | Indoor, outdoor, vehicle, phone line; noise type tagged |
| Device | Phone model class or studio microphone; sample rate and channels |
| File | Format | Contents |
|---|---|---|
| Audio | WAV / FLAC, 16 kHz+ (8 kHz telephony) | One file per utterance or session |
| Transcripts | JSONL / CSV | Verified reference text, timestamps on request |
| Metadata | CSV | Speaker band, dialect, district, condition, device |
| Consent records | Index file | Consent ID per speaker; signed forms held securely |
| Licence | TXT / PDF | Permitted uses agreed before recording |
Tell us the language, hours and conditions you need. We reply with a scope, timeline and quote.