Fish Audio, one of the bigger names in AI voice generation, just raised $52 million in seed funding to build out its models and push harder into enterprise deals. Gaming is explicitly one of the markets it's chasing, alongside video production, audiobooks, and customer-service bots. If you've ever considered using an AI voice for an NPC, a narrator, or dialogue in your own project, this is a company you've probably already looked at or will soon. It's also a company facing real, ongoing pressure over how it got some of its voices in the first place.

What Fish Audio Actually Does

The platform handles text-to-speech, speech recognition, and voice cloning, and it says it hosts more than two million user-uploaded voices. A clone can reportedly be built from as little as 10 seconds of reference audio, with controls for tone, emotion, whispering, and laughing layered on top. Its newer S2 model targets multi-speaker, multi-turn generation and can start producing audio in under 100 milliseconds under test conditions, which is the kind of latency that matters if you're building anything with real-time voiced characters. For game developers specifically, that combination, cheap, fast, expressive, is exactly what makes tools like this appealing. Recording and directing voice actors for every line of dialogue in a game is expensive and slow, especially for anyone working solo or on a small budget. AI voice generation promises to close that gap.

The Part That Should Give Any Dev Pause

British performers' union Equity formally demanded in June that Fish Audio remove unauthorized voice clones, after receiving complaints from 22 of its members. The union's allegation is that recordings, and the biometric data tied to them, were used to build models without informed consent. Individual cases add texture to that. Voice actor David Sheppard said 12 clones of his voice were removed following complaints, but pointed out that audio already generated from those models can keep circulating elsewhere even after the source model is taken down. Another performer, Faye Dicker, said her voice had been downloaded on the platform more than 900 times, and said she had no way of knowing what any of that generated content had actually been used for. A separate case involving a French voice actor illustrates the same gap: even after Fish Audio confirmed in writing that a model had been deactivated, new audio was reportedly still being generated from it the same day. Fish Audio has built a formal "Voice Ownership Dispute" process, where a claimant submits identity documents and reads a verification sentence aloud to help confirm ownership of a disputed voice. That process isn't automatic removal, though, claims get reviewed on their merits, and the burden sits with the performer to find the clone, prove it's theirs, and file the complaint in the first place. With a library north of two million voices, manual detection by rights holders alone is a genuinely hard problem.

Why This Isn't a Simple "AI Bad" Story

It's worth being fair here rather than piling on. Fish Audio does draw real lines: its policy bans cloning public figures, celebrities, or private individuals without documented authorization, and it prohibits impersonation and content designed to pass synthetic speech off as real. It also offers a legitimate licensed path for voice actors who want to work with the platform directly rather than have their voice scraped without consent. The problem isn't that rules don't exist. It's the gap between having a policy and actually enforcing it at the scale of millions of user uploads, and that gap is where real performers are getting hurt while the company keeps growing. Removing a listing and actually erasing someone's voice from a model turn out to be two different things, and that distinction matters more the bigger this technology gets. For any developer weighing whether to use AI voice tools in a project, this is exactly the kind of detail worth understanding before you commit to a platform, not after.

Sourcing Note

The funding figure and Fish Audio's own statements come from NewsDrum's original reporting and the company's public materials. The Equity complaint and individual voice actor accounts referenced here are independently corroborated by separate reporting from The Stage and the BBC/AI Commission, not solely from the original article or from Fish Audio itself. Would you use an AI-cloned voice in your own game, knowing what you know now? Let us know your thoughts in the comments. And if you've got a game you'd like us to review, send it over to editor@gamevibe.io.