Fish Audio is building a real-time, expressive voice platform that can clone a voice from a short clip in seconds, support 83+ languages, and offer more than 15,000 natural-language controls for emotion and pacing.

The platform provides text-to-speech, speech recognition, voice cloning, and long-form audio tools, hosting a massive library of over two million user-submitted voices and offering API access for developers and enterprises.

Its flagship model, S2.1 Pro, demonstrated strong performance in blind tests, with about two-thirds of listeners preferring its output over competitors.

The business mixes free/open access with paid enterprise plans, leveraging a rebuilt inference stack and FP8 GPU kernel library to keep the free tier affordable while serving commercial customers who need service-level guarantees.

Industry voices stress the importance of consent, transparent licensing, clear reporting and takedown workflows, and potential revenue sharing to build trust with creators.

Fish Audio operates in a competitive space with players like ElevenLabs and WellSaid, aiming to differentiate through fine-grained developer controls and cost-efficient model training.

The seed round features cross-border investors, signaling a trend of treating voice tech as an infrastructure layer and raising questions about governance, consent, and attribution for voice data.

Backers include Coreline Ventures, Capital Today, 359 Capital, and HF0 founder residency, highlighting strategic implications for a community-driven voice library and ongoing focus on consent and transparency.

The platform targets developers and regulated industries with HIPAA-compliant on-premises deployments, zero-data-retention policies, and plans to expand enterprise sales and API integrations with partners like Retell AI and LiveKit.

There is an ongoing governance dialogue over user-submitted voices, including a voice-ownership dispute process introduced recently and improvements in removal timelines after concerns about consent.

Open-weight models and a hosted API serve more than 8 million users and roughly $21 million in annual recurring revenue, supporting advances toward higher-quality models and broader enterprise sales.

The library now includes over 15,000 natural-language controls, serving both creative professionals and enterprises automating customer support and sales.