Fish Audio lands $50M seed funding to expand AI voice-generation platform

Fish Audio, a startup building AI voice-generation tools for creators and businesses, has secured $50 million in seed funding led by Coreline Ventures and Capital Today. The company, which has already reached $21 million in annual recurring revenue with over 8 million users across its open-source and hosted platforms, plans to use the capital to develop more sophisticated models and expand its enterprise offerings.

From open-source project to startup#

Founded by Shijia Liao, a former NVIDIA researcher, Fish Audio began as a frustrated engineer's side project. Liao built an expressive voice-generation model on a single GPU and released it publicly, frustrated by the flat, lifeless quality of existing synthetic voices. The Fish Speech repository on GitHub has accumulated over 31,000 stars and attracted indie game developers, video creators, and other builders looking for more natural-sounding voice options.

The company has rolled out five models over the past year—including four speech-generation tools and one speech-to-text system. While three of its speech models are available as open source, the latest S2.1 Pro version is limited to its paid API tier, signaling a shift toward monetization.

Tuning voices for different markets#

Fish Audio's platform includes roughly 15,000 fine-tuned natural language controls, allowing developers to shape tone, emotion, and delivery with precision. This approach appeals to different customer segments: gaming studios want expressive character voices, avatar platforms like HeyGen need photorealistic speech, and call-center automation providers want low-latency, conversational-sounding output.

The company offers tiered pricing for individual creators and teams, bundling generation minutes with voice-cloning features. For larger organizations, it provides dedicated enterprise APIs, already in use by companies like HeyGen, Sanas, and Plaud.

Addressing consent challenges#

Fish Audio's community-sourced model—compensating users whose voices train its models—hit a speed bump when creators discovered their voices had been uploaded without permission. Although the startup maintained a DMCA takedown process, removals were slow. CEO Rissa Cao announced that Fish Audio has now automated the takedown flow: creators can submit a voice sample or proof of ownership and have their voices removed within three minutes.