Buyers of call audio

Terms last checked against each company's own page: November 7, 2026

Five buyers publish terms that touch call archives. Labelbox Prism is the only one whose page names voice data, and it requires explicit, timestamped consent. micro1, Scale AI and Nyne take support records as text, and Defined.ai's catalog includes speech.

How do the published terms compare?

Each column is one buyer, quoted or closely paraphrased from its own page. Where a cell reads "Not published", the buyer was silent on the point; silence is not a no. We are independent of every company named, and none of them endorses this site.

Read on each company's own page, Nov. 7, 2026
Term	Labelbox Prism	Defined.ai	micro1	Scale AI	Nyne
What its page lists	Names long-running customer support histories, and voice and likeness where the people involved consented.	Sells speech among other data in its marketplace. What it accepts from suppliers is not spelled out.	Wants support and sales or CRM records, among many other kinds of company data.	Takes support records, plus chat and messaging histories.	Wants support tickets, service histories and CRM data.
Audio	Yes. Voice appears on its list, conditional on consent.	Speech is sold in its catalog; supplier intake is not specified.	No mention. The page talks about text, code, tables and workflows.	No mention. The page covers text, tables and workflow records.	No mention. The page lists records, tickets and spreadsheets.
Published pay	About $50K for a small targeted dataset at the low end, over $1M for the largest enterprise partnerships. Paid as an upfront fee with revenue sharing on top.	Not published.	Three published bands per company: $100K–$500K, $500K–$3M, and above $3M for highly unique data. Nothing per hour.	$10K to $1M+ per partnership, a figure Scale itself labels illustrative and which grows with delivery cadence. No guarantee.	Says deals commonly pass $100K, depending on what data is offered and how much.
Size or eligibility floor	No general size rule. For support logs it leans toward high-volume consumer businesses.	Applicants are vetted, and the page puts approvals at 30%. Sending its form means accepting supplier terms that contain a nondisclosure clause.	Wants 30 or more full-time staff and mostly English data; the US, UK and Canada are where its demand is strongest.	Prefers companies of 40 or more people, judged one case at a time.	A US payroll of 20 or more full-time W-2 employees, and three years operating.
De-identification	Prism cleans and strips personal data itself; location of that work is unstated.	Not published.	De-identifies on its side and lets the company review sample output.	Strips identities within the seller’s own systems before extraction; each data package needs sign-off.	States that data is de-identified ahead of release, without saying by whom.
Consent or exclusions	Voice or likeness is refused unless consent was explicit and timestamped.	Not published.	Not published.	Leaves out the identities of staff and customers.	Not published.
Payment timing	Claims under 21 days from the first conversation to money received.	Not published.	Full payment 30 days after acceptance per its process section; its FAQ ties the 30 days to invoicing instead.	Not published.	Not published.
Exclusivity and ownership	Exclusive unless negotiated otherwise; scope and sensitivity shape the terms.	Not published.	Claims no exclusive rights. Ownership of the data stays with the company.	Not published.	Not published.

Any license signed after one of our introductions may bring us a fee, paid by that buyer. Order and wording here ignore that: rows run from the most to the least direct mention of voice or call data.

What should a call archive owner read first?

Consent conditions. Prism excludes voice without explicit, timestamped consent. A recording notice is not the same thing, so ask Prism what evidence it accepts before you spend time on sizing.

Where de-identification happens. Scale AI strips identities while data is still on the seller's systems. micro1 and Prism do the work after receiving it, and their pages do not say where. For audio that may contain voiceprints, this difference matters.

Exclusivity. Prism licenses exclusively by default. That can rule out a second license for the same calls, so weigh it before signing.

Headcount. The three text-record buyers each set a staff minimum: Nyne the lowest at 20 US employees on W-2 (with three years trading), micro1 at 30, Scale AI at 40 as a preference.

Other channels that publish terms

General data marketplaces let a provider list a dataset and set its own price; they are not buyers. AWS Data Exchange, Snowflake Marketplace and Datarade publish provider terms of this kind (provider-set pricing, and in Datarade's case a marketplace commission), but they do not publish what a call archive would earn. For recorded calls, the consent and biometric questions are the same whichever channel you use.

Terms changelog

What changed on the buyer pages we track, newest first.

Date	Change
Nov. 7, 2026	Baseline: the five buyer pages in the table were read in full and their terms recorded.
Nov. 7, 2026	micro1’s headline band on its data page now tops out at "$3M+". When it changed is unknown.
Nov. 7, 2026	Left off for now: programs whose own pages could not be opened tonight. We add a buyer only after reading its terms on its own site.

Where these facts come from (8)

Prism, the licensing site run by Labelbox (seen Nov. 7, 2026)
Defined.ai, supplier partnerships (seen Nov. 7, 2026)
micro1, company data page (seen Nov. 7, 2026)
Scale AI, data partnership program (seen Nov. 7, 2026)
Nyne: selling company data (seen Nov. 7, 2026)
AWS Data Exchange provider financials (seen Nov. 7, 2026)
Snowflake provider listings pricing models (seen Nov. 7, 2026)
Datarade provider application page (seen Nov. 7, 2026)

Questions about buyers of call audio

Which buyer pages mention voice or call audio directly?

On Nov. 7, 2026, Labelbox Prism alone put voice on its list, and only where consent was explicit. Defined.ai's catalog includes speech, but its supplier page does not say what it accepts.

Do micro1, Scale AI or Nyne take call recordings?

Their pages list support records or support tickets, and describe text, tables and workflow data. None mentions audio. A transcript archive may be closer to what they describe; ask before assuming.

Why is there no per-hour price in this table?

Because none of these pages publishes one. The published figures are per company or per dataset. Anything quoted per hour elsewhere did not come from these pages.

Why are some well-known names missing?

We list a buyer only when we could read its own page and that page sets out terms for sellers. Anyone we could not check stays off the table until we can.

Is a high published range a sign a buyer pays more?

No. These are marketing ranges written to cover every kind of company data. Scale AI’s own page uses the word "illustrative" for its numbers. Actual licenses are often far lower.

Are you affiliated with these buyers?

No. We run independently, and none of the companies named on this page has endorsed us or holds any tie to us.