Fish Audio raises $52M seed, discloses $21M ARR and 8M users

A voice-model startup pulling $21 million in ARR before its Series A is the kind of number that either changes how the market prices open-weights AI or gets quietly walked back in six months, and Fish Audio just put itself in that position. Per TechCrunch's reporting, the company closed a $52 million seed round led by Coreline Ventures and Capital Today, and CEO Rissa Cao disclosed the ARR figure along with more than 8 million users across the open-source and hosted versions.

The part worth staring at is not the funding amount, because seeds have gotten silly across AI voice, but the shape of the business Cao is describing. Fish Audio has released five models, three of them open-sourced, and its Fish Speech repository has more than 31,000 GitHub stars. Sitting on top of that community layer is a paid API and a set of enterprise customers TechCrunch names, including HeyGen for AI avatars, Sanas, and LiveKit for voice agents. Cao's own framing is that different customers want different things from a voice, from realism for AI avatars to 'more natural-sounding and low-latency voices' for voice agents.

The competitive set she is walking into is crowded, and TechCrunch names it plainly: ElevenLabs, WellSaid, Cartesia, Speechify, Async and Krisp. In that context the interesting operational detail is the consent story. Fish Audio has automated its takedown process so that a creator submitting a short voice sample or a contract can have an uploaded clone pulled from the platform in under three minutes. In a category where enterprise deals stall on legal risk and reputational exposure, a same-hour takedown SLA is not window dressing.

The honest caveat is that most of what the reporting gives you is the company's own telling. The ARR is disclosed by Cao, not audited. The round is called a seed but no valuation is attached, and the split between free open-source users and paying seats inside the 8 million figure is not broken out. Nor does the article say how many enterprise customers make up the revenue, which matters when you are trying to figure out concentration risk.

What is worth watching is whether an open-weights base plus a fast consent mechanism is enough of a wedge against ElevenLabs and Cartesia to matter over the next year. For teams already building voice agents or AI avatars, a credible second-source vendor with a public takedown SLA is useful leverage even if you stay on your current provider.