Fish Audio Secures $50M to Scale AI Voice Models

Founded by former NVIDIA researcher Shijia Liao, the startup initially gained traction through an open-source speech repository that now boasts over 31,000 GitHub stars. Fish Audio differentiates itself by offering more than 15,000 natural language controls, allowing developers and gaming studios to steer voice output with precision. While the company maintains open-source versions of its earlier models, its latest S2.1 Pro iteration remains locked behind a paid API, serving high-profile corporate partners such as HeyGen and LiveKit.

Rapid growth has brought scrutiny regarding intellectual property. After facing allegations that unauthorized voice samples were uploaded to the platform, CEO Rissa Cao stated the company has implemented an automated takedown process, allowing creators to remove content within three minutes. Despite these safeguards, the challenge of preventing unauthorized uploads remains a hurdle for a platform built on community-driven data. Investors like Oskue Honda emphasize that the long-term viability of the model depends on prioritizing consent and transparent revenue-sharing with the creators whose voices power the technology.

Competition in the synthetic speech sector is intensifying, with ElevenLabs, Cartesia, and Speechify vying for the same market share. To maintain its edge, Fish Audio plans to launch an audio understanding model and a speech-to-speech system later this year. According to Rico Mallozzi of 359 Capital, the startup's ability to produce human-like quality with relatively lean resource consumption is a critical advantage against larger, well-capitalized AI labs.