🤓 Latest Submissions
VoxBox
AI voice cloning makes it trivial to generate speech in someone's voice — but without a consent layer, that's a liability for the voice actor and a legal grey area for whoever uses it. VoxBox makes consent explicit, scoped, and enforced in software instead of relying on a one-time waiver. Voice actors record a short sample and check off exactly which categories of use they're pre-approving — ads, political content, adult content, medical claims, financial advice, and more. That's not a formality, it's an allow-list the platform actually enforces. When a renter submits a script, an LLM (Gemma, via Fireworks AI) classifies which of those categories the script touches. A deterministic policy layer then routes it: if every flag is covered by the actor's declared consent, the script is generated instantly. If it raises something outside consent, it goes to the actor's approval queue as a visible, explainable decision —never a black box. Two categories (hate speech/violence, impersonation) are hard-blocked outright if not declared, no human needed. Voices are priced per 100 words generated, not as a flat rental fee — renting a voice just unlocks it. The real cost is transparent and shown live as the renter types. Actors get a dashboard of usage and revenue per voice. Built with FastAPI, Chatterbox Turbo TTS for cloning (running on AMD GPU instances on DigitalOcean yet to try some AWS instances) and Gemma4 running on Fireworks for screening and challenge responses.
13 Jul 2026
