Fish Audio Raises $52 Million to Build AI Voice Models for Creators and Enterprises

Humble Beginnings, Extraordinary Takeoff

Fish Audio started as a small personal project launched by former Nvidia researcher Shijia Liu, who felt genuine frustration with the limited expressiveness of synthetic voices available on the market. In response, he decided to train a voice generation model using a single GPU and released it publicly as an open-source project. Today, the Fish Speech repository on GitHub has surpassed 31,000 stars and is relied upon by independent developers, video game designers, and content creators around the world.

Substantial Funding Fuels Growth

Fish Audio, headquartered in Palo Alto, California, announced the completion of a seed funding round totaling $52 million, led by Coreline Ventures and Capital Today, with participation from a number of other investors including 359 Capital, Parable, HF0, and others. This funding comes at a time of strong momentum for the company, which boasts a user base of over 8 million people and an estimated annual recurring revenue of $21 million.

A Comprehensive Voice Library for Diverse Needs

Fish Audio's philosophy centers on delivering voice solutions adaptable to a wide range of use cases, through a library of more than 15,000 natural language control tools. The company has launched five models over the past year:

• Four speech generation models, three of which are open-source.
• One speech-to-text model.
• The latest model, S2.1 Pro, available exclusively through a paid API.

The company offers monthly plans for creators and teams that allow generation of a set number of audio minutes with voice cloning capabilities, as well as an enterprise tier already in use by prominent companies such as HeyGen and Sanas.

Diverse Use Cases Create a Competitive Edge

CEO and co-founder Risa Cao explains that each organization has a distinct need: HeyGen, which integrates voices into AI avatars, prioritizes realism; game studios need more expressive and dynamic character voices; while voice agent companies like LiveKit prefer natural, low-latency voices that closely mimic human conversational style.

The Consent Challenge and Protecting Creators' Voices

Fish Audio built its voice library through a community-driven model that invites users to contribute their voices in exchange for financial compensation when those voices are used. However, this approach led to disputes when some creators claimed their voices had been uploaded without their knowledge or consent. The company responded by developing an automated removal mechanism that allows voice owners to submit a voice sample or proof of ownership and complete the removal process in under three minutes.

Yet the fundamental challenge remains: nothing prevents someone from uploading another person's voice without their knowledge, meaning voice protection ultimately depends on owners being vigilant themselves. Osuki Honda, Partner at Coreline Ventures, believes the sustainability of this community-driven model hinges on creator trust in the platform, and that the industry as a whole must adopt clear standards encompassing voice ownership verification, transparent licensing, rapid reporting mechanisms, and revenue-sharing models for creators.

An Ambitious Roadmap in a Competitive Market

Fish Audio plans to launch a voice understanding model and a speech-to-speech model during the current year, in a market crowded with prominent players such as ElevenLabs, Cartesia, Speechify, and others. Rico Mallozzi, Partner at 359 Capital, argues that Fish Audio's ability to build sophisticated models with a lean team and low training costs represents a genuine technical advantage — one that positions the company to compete even against well-funded major AI laboratories.

✦ بقلم فريق دروب أيديا