Your voice is an asset you own. Voice Vault lets voice actors and everyday people license their voice for AI with clear, revocable consent, get paid every time it's used, and fight back when someone clones it without asking.
2. Why people can't resist it
Money while you sleep. Watching USDC tick in each time an audiobook chapter uses your voice is deeply compelling.
Protection from a real fear. Everyone has heard of voice clone scams. A free "Is this my voice?" checker brings people in.
Control. Toggles for what your voice may and may not say turn a scary technology into something you run.
3. How it works for real
Login: email or phone, a passkey smart wallet is created behind the scenes (Coinbase Smart Wallet or Privy). Gas is sponsored on Base. Payouts are USDC, cashed out to mobile money (MTN MoMo, M-Pesa), bank or card.
Identity and liveness: a KYC partner checks ID and face; a random liveness phrase is spoken on camera and matched to the sample with a speaker verification model (ECAPA-TDNN style embeddings).
Voice model: 3 to 10 minutes of reading is used to build a voice with a commercial TTS provider (ElevenLabs Professional Voice Cloning, Cartesia, or an open model like XTTS or F5-TTS self-hosted). The model stays in Voice Vault's custody; buyers never download it.
Consent on-chain: a license record (voice ID hash, allowed uses, price, expiry, revocation flag) is written to a contract on Base. Every generation checks it first.
Use check: each buyer script is screened by a fast cheap classifier (Gemini Flash or DeepSeek via OpenRouter) against the license: political, adult, hateful or impersonation content is blocked. A frontier LLM (Claude) handles edge-case appeals.
Watermark and log: every clip carries an inaudible watermark (AudioSeal style) with the generation ID, and a hash of each clip is logged on Base.
Pay: buyers pay per minute in USDC with x402, so apps and agents can call the voice API and pay per request. The creator's share streams to their wallet instantly.
Detect: uploaded clips are checked for the watermark and compared by speaker embedding; matches without a license trigger a takedown pack.
4. Cost per use
TTS: about $0.05 to $0.30 per minute of audio depending on provider. Classifier check: under $0.001. Watermark and log: about $0.001. At a $1.20 per minute buyer price, the creator gets 70% ($0.84), Voice Vault keeps about $0.20 after costs.
Free: record, verify, list one voice, run 10 detection checks a month. Pro ($9 a month): unlimited detection, auto takedown filing, exclusivity deals and custom licenses. Buyers pay per minute, no subscription needed.
5. Growth loop
Each listed creator shares their "Hear me in your app" card with their audience. Every clip a buyer publishes carries a "Voiced with Voice Vault" credit link. The free clone checker pulls in people who were cloned, and many of them list their voice. Every Monday a weekly voice report shows earnings by buyer, a running earnings chart and any new clone matches with one-tap takedowns, so creators open the app every week.
6. Risks and guardrails
Consent: only the verified owner can list a voice. Liveness and ID checks block uploading someone else's recordings.
No voices of others without proof: estates and agencies must show written authority; celebrity and public figure lookalikes are rejected.
Banned uses for everyone: fraud, impersonating a real person, political persuasion without a toggle, sexual content without a toggle, hate and harassment.
Watermarking: every clip is traceable to a buyer and a license.
Revocation: revoke stops new generations instantly; already delivered clips stay licensed for their stated use only.
Law: follow right-of-publicity and AI disclosure rules (EU AI Act labelling, US state voice laws).
7. MVP build order
Email login with a smart wallet, record a 3 minute sample in the browser, liveness phrase check.
Create a voice with a TTS provider API, and a license toggles screen saved to a Base Sepolia contract.
Buyer page: preview, script check against the license, pay per minute in test USDC.
Watermark each output and show the usage log with live earnings.
Clone checker with speaker embeddings and a takedown request template.