Introducing the British Voices Bank: open voice data for the languages the world forgot.
On air Voice improv, for real money

Get paid to play.
Teach AI to talk like home.

Accent Studio is a voice game. Get dealt a character, improvise a scene with a stranger, and get paid while you build the most human voice datasets on earth.

See how it plays → I'm a lab
Pidgin · Yoruba · Swahili · Zulu + 8 more Zero PII by design Cash out via M-Pesa · Paystack · USDC
rated 5.00
Platinum ×1.2
🎭 Blind Duet
01
01 · The games

One studio, two games.
Each one mints a different kind of data.

Pick your energy. A quick two player rally you can play on a bus, or a full live scene with a castmate. Both pay, and every transcript is verified afterwards by a trained linguist, not by another player.

Async · 2 playersASR · transcription

Ping-Pong

Cards arrive in your language: the situation, and who you are in it. You say each one your way. Your partner rates the take on tone, prompt, mood and clarity. Below 4? You say it again. Every clip is verified by a linguist before it ships.

Your card: the situation, and who you are in it.
  • Turn based and async, play anytime on any phone
  • Peer rated on 4 axes, redo below 4
  • Produces clean, isolated single speaker turns
Live · 2 playersvoice-to-voice · dual channel

Live Arena

Two players hold a flowing, natural conversation. A market haggle, a support call, a complaint that escalates. The scene card is in your language: the situation, and who each of you is in it. The result is the rich, emotional audio voice agents are starved for, verified afterwards by a trained linguist.

You are dealt a persona, in your language.
  • Real turn taking, real interruptions, real emotion
  • A trained linguist verifies the transcript and fixes slang afterwards
  • Produces dual channel, latency mapped conversation
02
02 · In your hand

The whole studio lives on your phone.

Play scenes, hear takes, rate your partners and watch your earnings build, all from the device you already carry. Built mobile first, so anyone with a phone and a story can join in.

0 minVoice note turn
0 minFull Arena scene
0+Dialects on day one
03
03 · The mechanics

We didn't gamify a chore.
We built a game that happens to mint gold.

01

Blind Duet

No names, no photos. You are always paired with a stranger, so every scene has mystique and no one can rig a take.

fun + fraud proof pairing
02

The Twist

Mid scene, a curveball drops for both of you. Reject the first apology. Switch to Pidgin now. It keeps every take gloriously unscripted.

fun + natural, unscripted audio
03

Chemistry

A live meter fills while your scene flows. High Chemistry means a bigger payout and a shared Standing Ovation. Instant dopamine, real quality signal.

fun + live quality scoring
04
04 · Hear a scene

Press play. This is a flowing take.

Real scenes from the game: dual channel, transcribed, emotion tagged, latency mapped. English source on every line.

Nigerian Pidgin / English
2 speakers · 24 kHz · dual channel
Recorded by the founding cast, unedited. Two speakers, one phone each, one take.
05
05 · Earn

Play well. Rank up. Cash out.

A published base rate for your language, shown before every session. A quality multiplier applied after a linguist verifies it. Real money on the rails you actually use.

$8 – $30 per speaker, per verified hour of session audio Standard base rate $16 per speaker per hour. Scarce registers and commissioned specialist domains pay more.
Peer QA multiplier · applied after verification

For labs

Every scene played becomes the exact row a lab pays for.

Clean, multi turn, non Western conversational audio. Born digital, editor verified, consent signed, ready for a training pipeline on day one.

Explore the data →
WAV · dual channel · 24 kHz
JSONL · turn level metadata
Parquet · searchable index
Aligned · word level EN ↔ target
Consent · cryptographically logged
Safety

No real names. No account numbers.
No privacy risk.

🎭

Fiction only

Every scene is a synthetic scenario. The "account number" you confirm is a stage prop, never real data.

🕶️

Anonymous pairing

Partners never see your name, face or handle. Pairing is engineered so acquaintances never match.

🔏

Consent, signed

Your agreement is time stamped and hash anchored. Voice is biometric data and we treat it that way. Read it below before you sign.

🧹

Auto redaction

Anything that looks like real PII is caught at capture and replaced with synthetic tokens before storage.

Read the agreement before you sign it.

The full Contributor Agreement every player signs: plain-language summary first, then every clause, including what buyers may never do and how you withdraw.

v1.2 · effective 12 September 2026 · 11 pages · SHA-256 48ff6cdd…97d1 printed in full on the Labs page
Download the agreement ↓

Honest play, and what happens if a session is questioned.

You agree to three things: your own live voice, recorded in the session, in a language you actually speak; the scene as briefed; and honest ratings of your partner. Your partner's voice stays inside the app too: you listen to rate and reply, never download, keep or sell it, and the same rule protects yours. Dishonesty means things like synthetic, pre-recorded or replayed audio, someone else recording as you, more than one account, trading ratings, or filler content across sessions. If a session is flagged, a person reviews it before anything happens, you are told what was flagged and why, and you have 7 days to respond. Only confirmed dishonesty forfeits pay, and only for the sessions involved. Verified earnings from other sessions are still paid. Low quality is never treated as dishonesty; it goes to a redo or the quality floor. Clause 15 of the agreement is the full text.

Your voice is biometric data, so we never disguise it and never expose it. Recordings are stored under a random ID. Your name, ID and payout details sit in a separate vault that buyers never receive. Every buyer signs that they will not try to identify you, link you to other data, or clone your voice. You can request a copy of your data or delete your identity from Settings at any time. Details in our privacy policy.

06
06 · Good to know

Questions, answered.

How do I actually get paid?
You earn a published base rate per speaker per verified hour of session audio, shown before you start, multiplied by your QA tier after a linguist verifies the session. The current rates and tiers are in the Earn section above. Cash out to M-Pesa or Paystack/Flutterwave from $5, Stripe or PayPal from $10, USDC from $2. Payouts land within 7 business days of your tally closing.
Do I need a good microphone or fast internet?
No. A mid range phone in a quiet room is the target setup. Ping-Pong is fully async, so patchy internet is fine. Live Arena needs a stable connection for the session length, about 5 minutes.
How can my pay change after a session?
Who hears my voice, and what does my partner see about me?
Your partner hears your live voice for the length of the session, and you hear theirs. They never get your name, contact details, handle or a copy of the recording. Our trained linguists listen afterwards to verify the transcript; they see a random speaker ID, never who you are. Buyers receive pseudonymous datasets and are contractually barred from trying to identify anyone.
Do I need to be able to read and write my language?
Scene cards are a few short lines in your language: the situation, and who you are in it. Reading those is all it takes. You never read lines aloud, nobody checks your spelling, and a linguist writes the transcript.
What happens if a session is flagged, and can I respond?
A flag changes nothing on its own. A person reviews the session, you are told what was flagged and why in the app and by email, and you have 7 days to respond before any decision. Only confirmed dishonesty forfeits pay, and only for the sessions involved; everything else you earned is still paid. A poor-quality session is never treated as dishonesty. It is clause 15 of the agreement.
Who do I play with?
Strangers, always. Blind Duet pairing is anonymous by design and engineered so people who know each other never match. It keeps scenes fresh and the data honest.
What data is collected about me?
Your voice recordings from scenes, your ratings, and payout details for your chosen rail. Your identity is stored in an isolated encrypted vault, decoupled from your audio, which is indexed only by an anonymous ID. Labs never see who you are.
I'm a lab. How do I license a corpus?
Head to the labs page for schema, sample bundles and delivery formats, or write to us directly. Pilot corpora ship in weeks, not quarters.
The waitlist

The stage is almost open.

First seats go to the loudest voices. Apply now and we'll call you up when your language goes live.

Apply as a player →

Four minutes on your phone: the languages you speak, a short voice sample in each one you want to work in, and how you'd like to be paid later. No account needed.

Launching first in Lagos · Nairobi · Accra · Joburg