Sell Voice Data
48 kHz 24-bit mono 1 speaker 312 hours
What a buyer's spec sheet looks like (illustrative)
Buyers price voice recordings from the spec sheet first: sample rate, speakers, hours and signed releases decide who will take a set before anyone listens.
Get a quoteWhat makes voice data sellable?
Nine lines, read top to bottom. The right-hand column is the illustrative set from the line above; swap in your own values.
| Field | Example | Why buyers check it |
|---|---|---|
| Sample rate | 48 kHz | Detail lost at recording time cannot be added back, so a higher native rate keeps more uses open. |
| Bit depth | 24-bit | Lossy files (MP3, AAC) have already discarded detail; buyers often treat them as a lower grade. |
| Channels | mono | Per-speaker tracks save the buyer the work of splitting and labeling voices. |
| Speakers | 1 speaker | One voice for many hours suits voice cloning; many voices suit recognition. The mix decides who buys. |
| Accents and languages | US English | Under-represented accents and languages are scarcer than standard US English. |
| Hours | 312 hours | Volume matters, but only once the rest of the sheet passes. |
| Transcripts | human-checked | Aligned, checked text makes audio usable for training without a labeling project first. |
| Environment | treated room | Consistent, quiet audio is cheaper to use. Mixed conditions should be described as separate sets. |
| Releases | signed, all speakers | No release, no deal: most buyers will not take a voice they cannot prove consent for. |
Marketplace or direct buyer: what is the difference?
A marketplace lists your dataset for many buyers and takes a cut. A direct buyer licenses it once, usually on terms it sets. Both routes, side by side, from the companies' own pages:
| Marketplace listing | Direct license | |
|---|---|---|
| Who sets the price | You do, on your listing (AWS Data Exchange, Datarade) | Set in the license; Labelbox Prism quotes a published floor near $50K a dataset |
| What it costs you | AWS: 1.5% to 3% listing fee. Datarade: 15% to 30% commission plus a yearly plan fee of $0, $6,000 or $12,000 (Gold is custom) | No seller fee published |
| How money arrives | As buyers pay: AWS disburses monthly, after it collects from the subscriber | Labelbox Prism: money on signing, a revenue share afterwards, and a first wire inside 21 days by its account |
| Exclusivity | Not stated on the pages read | Labelbox Prism defaults to exclusive licenses |
| Who can sell | AWS: registered Marketplace sellers. Datarade: registered businesses only | No size rule published; consent for every voice required |
| Best fit | A set you want to sell many times, at a price you control | A distinctive set one buyer may want to own |
Disclosure: a company in this list can owe us a referral payment when one of our introductions becomes a signed license. Payment has no bearing on the order (alphabetical) or on the wording.
Who buys voice data, and what do they pay per hour?
Companies with a published route for voice recordings, alphabetically. On the pages read 7 November 2026, none published a price per hour of audio.
| Company | Route | Published pay | Per hour |
|---|---|---|---|
| AWS Data Exchange | Marketplace listing | The provider sets its own price; no payout figures published | not published |
| Datarade | Marketplace listing | The provider sets its own price | not published |
| Defined.ai | Marketplace listing | not published | not published |
| ElevenLabs | Usage royalty | Earnings based on use by paid users (free users excluded); no per-character rate published | not published |
| Labelbox | Direct license | From around $50K for a small targeted dataset to $1M+ for a large enterprise deal; a license fee up front, with revenue sharing on downstream use after that | not published |
Disclosure: a company in this list can owe us a referral payment when one of our introductions becomes a signed license. Payment has no bearing on the order (alphabetical) or on the wording.
Where these facts come from
- AWS Data Exchange, provider financials documentation (read on 7 November 2026)
- AWS Marketplace, listing fee schedule (read on 7 November 2026)
- Datarade, provider plans on its apply page (read on 7 November 2026)
- Defined.ai, supplier partnerships overview (read on 7 November 2026)
- ElevenLabs, help article on voice actor payouts (read on 7 November 2026)
- ElevenLabs, Voice Library payouts docs (read on 7 November 2026)
- Labelbox, Prism licensing site (read on 7 November 2026)
Fees, payment timing and rights for each buyer or run the hours calculator
How does a voice data deal run?
Inventory
Count hours of speech, speakers and formats, and split the archive into consistent sets.
Clear rights
Match every speaker to a release and set aside what you cannot clear.
Describe, then sample
Buyers read the spec first. Samples go only under the buyer's own agreement, with personal details removed.
License
The license sets scope, exclusivity, price and payment; your lawyer reads it before you sign.
Deliver and get paid
Delivery and payment follow each buyer's terms. Labelbox Prism, for one, pairs a signing payment with an ongoing revenue share.
Questions about selling voice data
- Can I sell voice recordings for AI training?
- Yes, if you own the recordings and every speaker signed a release that covers licensing for AI training. Buyers such as Labelbox Prism state that they exclude voice without explicit, timestamped consent.
- How much is an hour of voice data worth?
- No buyer we track published a per-hour price for voice recordings on 7 November 2026. Labelbox Prism prices whole datasets, with a floor near $50K. Treat any per-hour figure without a named source as a guess.
- Do I need transcripts?
- No, but they help. Human-checked transcripts with timestamps make audio usable without a labeling project, which widens the list of interested buyers.
- Can I sell a podcast with guests?
- Only the parts you can clear. Every guest needs a release that covers AI training; episodes with guests you cannot reach are usually left out.
- Is my voice biometric data?
- It can be. Illinois's BIPA, and laws in Texas and Washington, can treat voiceprints as biometric identifiers. Ask a lawyer before selling recordings that could be used to recognize or clone a speaker.
- Do you buy the recordings?
- No. Sell Voice Data never receives or hears any audio. Our job is matching your description to buyers' stated conditions and making an introduction once you approve it.
- What about call center recordings?
- Recorded customer calls raise different consent questions from studio and podcast audio, and this site does not cover them.
Get a quote
Give us the company ZIP, an address for our reply, then tap through some questions on the audio. Our reply names the buyers whose own published conditions your recordings appear to satisfy.
Describe the recordings in words. Audio files, transcripts, samples and download links should never be sent here, and we would delete them unread.
Sell Voice Data introduces sellers to buyers; a buyer you agree to meet can pay us a referral fee once a license is signed.
Received. Here is what happens next.
- Your answers are compared with the conditions each voice data buyer publishes on its own site.
- An email follows listing the companies that match, with links to their terms. A poor fit gets an honest "none yet".
- Your name stays with us unless you answer that email approving a particular buyer.
Leave the audio where it is stored. Any buyer wanting samples asks you directly, under its own agreement.