Fish Audio raises $52M seed to build AI voice models for creators and enterprises
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Fish Audio raised a $52 million seed round led by Coreline Ventures and Capital Today, with participation from 359 Capital, Parable, Play Time, Alphalist Partners, Bayhouse Ventures, Carya Venture Partners, and HF0.
- Fish Audio has reached 8 million users and $21 million in annual recurring revenue, offering more than 15,000 natural language controls and having launched five models in the past year — three open-sourced plus the paid-only S2.1 Pro.
- Fish Speech, the startup's open-source GitHub repository, has surpassed 31,000 stars and was originally built by former NVIDIA researcher Shijia Liao, who trained the first voice generation model on a single GPU; enterprises including HeyGen, Sanas, and LiveKit now use Fish Audio's APIs.
- Fish Audio faced a consent controversy after creators alleged their voices were uploaded without permission; CEO Rissa Cao said the startup has now automated its DMCA takedown process to under three minutes, though the article notes unauthorized uploads can still go undetected until an artist files a claim.
- Coreline Ventures partner Osuke Honda argued that a community-driven model only works if creators trust the platform, calling for verified voice ownership, clear licensing, easy reporting, and revenue-sharing models built into the product from the start.
- Fish Audio competes against ElevenLabs, WellSaid, Cartesia, Speechify, Async, and Krisp; 359 Capital partner Rico Mallozzi credited the startup's fine-grained developer controls and cost-efficient model training for closing the gap on better-funded AI labs.
- Fish Audio plans to release an audio understanding model this year and is building a speech-to-speech model.
Why it matters: A seed-stage AI voice company landing $52 million while already sitting on $21 million in ARR and 8 million users signals outsized investor appetite for voice infrastructure outside the headline names — but the consent controversy is a structural trust liability that automated DMCA takedowns only partially address, since unauthorized uploads go undetected until an artist notices.
