Consented people, known places, metadata you can filter on
AI training data collection in India is mostly a field problem, not a software one. A speech model needs Bhojpuri speakers from Siwan and Awadhi speakers from Ayodhya, recorded in real rooms, with consent a lawyer would accept. Bharat Survey brings the part survey agencies already do well: recruiting specific people in specific districts, briefing them in their language, recording consent, and attaching verified metadata to every file. It is a project service scoped with you, for AI teams, research labs and product companies building for Indian users.
Hello, I am Ram Kumar. This survey is run by Sample Organisation. Purpose: to learn about amenities in your area. It is voluntary, and you can say no at any time.
Crowd platforms are good at volume from people already online. They are weak at reaching older speakers, rural women, low-literacy contributors and dialects that are rarely typed. Those are the gaps that make a model fail in a district office or a village shop. A field team that already knocks on doors for household surveys can reach them, explain the task face to face, and record consent in the language the contributor speaks.
Read speech from prompts, spontaneous speech on set topics, conversational pairs, photographs of objects, documents or scenes you specify, and video when the client supplies or approves the capture hardware. Every item is linked to a contributor record holding consent, demographic fields you define and the capture location at district level.
We do not scrape, buy or reuse recordings from other projects. We do not annotate or transcribe at scale in-house; transcription can be arranged with a partner when scoped. We do not collect biometric identifiers for identification purposes.
Formats differ in how much hardware and supervision they need. Here is what each involves.
Hello, I am Ram Kumar. This survey is run by Sample Organisation. Purpose: to learn about amenities in your area. It is voluntary, and you can say no at any time.
| Need | Field team | Online crowd platform |
|---|---|---|
| Rural and low-literacy speakers | ||
| Verified district of contributor | ||
| Fast volume from urban smartphone users | ||
| Face-to-face consent explanation | ||
| Lowest cost per clip | ||
| Controlled recording conditions |
Many projects use both: a crowd platform for urban volume, field teams for the hard-to-reach quotas.
A voice recording or a face in a video is personal data under the Digital Personal Data Protection Act, 2023. The client building the model is normally the data fiduciary and decides the purpose; Bharat Survey collects on its instructions. The consent notice must say what the data will be used for, and contributors must be able to withdraw.
Contributor ID, language of the notice, date, the licence version agreed, payment acknowledgement where contributors are compensated, and any restrictions they chose. If a contributor withdraws, the ledger shows which files to remove.
Market and customer research for companies.
Learn moreVoice-based surveys over the phone.
Learn moreFace-to-face consumer interviews across India.
Learn moreReaching villages and small towns.
Learn moreRepeated waves with consistent samples.
Learn moreStraight answers to the questions people ask most about this topic.
Language coverage is scoped per project. We recruit through field networks in the districts where a language or dialect is spoken, so feasibility depends on the dialect and quota. Tell us the languages and speaker profile; the pilot proves whether we can meet the quota before volume collection begins.
Ownership and licence are defined in the contract and in the consent terms each contributor agrees to. Normally the client receives the rights contributors grant. We delete our copies when you instruct us and do not reuse contributors or files for other projects.
Compensation is common in speech and video projects and is agreed with the client up front. Where contributors are paid, the payment and its terms are recorded in the consent ledger so it is clear the contribution was voluntary and informed.
Only with wearable hardware supplied by you or an approved partner, written consent from the wearer, and rules for bystanders and private spaces agreed before capture. We manage recruitment, briefing, sessions and uploads. We do not start video capture until the pilot shows the consent process works.
Share languages, formats, quotas and your consent terms. We reply with a pilot plan and feasibility notes.
