Born from the open-source Fish Speech project, Fish Audio has secured a $52 million seed round after reaching $21 million ARR, betting that open models, self-hosting and free APIs can challenge proprietary voice AI leaders.
Fish Audio, a commercial AI voice company that originated from the open-source Fish Speech project, has raised a $52 million seed round led by Coreline Ventures and Capital Today, marking one of the strongest examples of an open-source AI project evolving into a venture-backed business. The company had already reached $21 million in annual recurring revenue (ARR) and 8 million users before taking its first institutional funding.
Alongside the funding, Fish Audio launched its S2.1 Pro voice model, which supports text-to-speech, voice cloning, voice agents and 83 languages. It has also made the model available through a free API under a fair-use policy to encourage developer adoption.
Founded by former Nvidia researcher and chief scientist Shijia Liao, Fish Audio traces its roots to the open-source Fish Speech project, with technical work published on arXiv and GitHub. Its March release included open weights, fine-tuning code, an SGLang-based streaming inference engine and self-hosting support, targeting developers and organisations seeking greater control over voice AI deployments.
According to Artificial Analysis, S2.1 Pro achieved an Elo score of 1,153, ranked 13th among text-to-speech models, processed 56.3 characters per second and supports natural-language emotion and prosody control. Fish Audio also said its own blind listening tests favoured S2 Pro over models from ElevenLabs, Inworld and MiniMax, although the evaluation was conducted internally.
Fish Audio now enters a competitive market led by ElevenLabs and Cartesia, but its combination of open-source foundations, developer-focused tooling and proven commercial traction sets it apart.
