Speech data from India's under-served languages
LootBeez runs a network of native speakers across Bihar, Jharkhand, Uttar Pradesh, West Bengal, Chhattisgarh and Nepal. We design, collect, verify and deliver custom speech datasets — with documented consent for every speaker.
Data types
Read speech
Your scripts or ours, read aloud by many speakers. Ideal for ASR training and TTS evaluation. Typical clip length 2–15 seconds.
Voice commands & intents
Wake words, assistant commands, IVR phrases and domain intents (agriculture, health, banking), spoken naturally with variations.
Spontaneous speech
Image-prompted and topic-prompted natural speech, 20–120 seconds, capturing real vocabulary, code-mixing and accent.
Transcription
Native-script transcription by speakers of the same dialect, followed by human verification and correction.
Custom metadata
Age band, gender, state, district and dialect for every pseudonymous speaker. Additional fields on request.
Evaluation sets
Small, carefully balanced held-out sets for benchmarking your models in low-resource dialects.
From brief to delivery
Scope
We agree languages, hours, demographics, prompts, file specification, timeline and price in a written order.
Pilot
A small pilot (typically 2–5 hours) to validate quality and spec before full collection begins.
Collect & verify
Trained contributors record on their phones. Every item passes automatic checks and human review. You track progress on a private dashboard.
Deliver
WAV files with CSV/JSON metadata, delivered in milestones by secure link, with a consent and QC report.
Default technical spec
Everything below can be customised in your order.
| Audio format | WAV, PCM 16-bit, mono, 16 kHz (other rates on request) |
|---|---|
| Clip length | Read speech 2–15 s; spontaneous 20–120 s; set per project |
| Automatic checks | Minimum loudness, silence ratio, clipping, duration limits, exact-duplicate detection |
| Human QC | Every delivered item approved by a trained native-speaker reviewer; reviewer cannot review own work |
| Speaker metadata | Pseudonymous speaker ID, age band, gender, state, district, dialect |
| Speaker limits | Configurable cap per speaker per project to guarantee speaker diversity |
| Consent | Versioned, time-stamped contributor consent and assignment of rights, aligned with India's DPDP Act 2023 |
| Delivery | Milestone deliveries with metadata.csv / metadata.json and a delivery report |
| Progress tracking | Private read-only dashboard link: approved hours, demographics, daily progress |
Reviewing us as a vendor? Our Trust & security overview covers consent, privacy, security and QC in detail.
Where we collect
Need another dialect from these regions? Ask us — we recruit through local coordinators.
Request a sample dataset
Tell us what you need. We reply within 2 working days with a free sample, a rate card and a proposed timeline.
- Pricing per approved hour, agreed upfront
- Milestone billing with advance and balance on delivery
- Replacement of any item that fails the agreed spec — see Refund Policy
- NDA available on request
Prefer email? Write to business@lootbeez.shop.