Fish Audio raises $50M seed to build AI voice models for creators and enterprises

The marketplace for AI-generated dependable models is massive. Creative usage cases necessitate AI dependable models to beryllium much expressive, portion enterprises looking to automate lawsuit enactment and income ops request them to beryllium much steerable.

Palo Alto-based Fish Audio wants to cater to each of those usage cases with its room of much than 15,000 earthy connection controls. Since launching past year, the startup contiguous has much than 8 cardinal radical utilizing the open-source oregon hosted versions of its models, and present generates yearly recurring gross of $21 million.

To proceed gathering connected that traction, the startup connected Tuesday said it has raised $50 cardinal successful a effect circular that was led by Coreline Ventures and Capital Today. The backing besides saw information from 359 Capital, Parable, Play Time, Alphalist Partners, Bayhouse Ventures, Carya Venture Partners, and HF0.

Fish Audio started arsenic a tiny task by erstwhile NVIDIA researcher Shijia Liao, who, frustrated by non-expressive synthetic voices disposable connected the market, trained a dependable procreation exemplary connected a azygous GPU, which helium open-sourced. The Fish Speech repository connected GitHub present has much than 31,000 stars, and is utilized by indie developers, video crippled designers, and creators.

The institution has launched 5 models successful the past year: 4 code procreation models and 1 speech-to-text model. It has open-sourced 3 of its code procreation models, but its latest S2.1 Pro exemplary is disposable lone done its paid API.

Fish Audio offers paid monthly plans suited for creators and teams that unlock a acceptable fig of minutes of generation, positive dependable cloning features. The institution besides offers an endeavor mentation of its APIs and platform, and says organizations similar HeyGen, Sanas and Plaud are already utilizing it.

“Every endeavor has antithetic usage cases and antithetic preferences. For example, companies similar HeyGen, which usage our voices to powerfulness AI avatars, privation realism successful voices; a gaming workplace would privation expressive dependable for their characters; and dependable cause companies similar LiveKit privation much natural-sounding and low-latency voices that are expressive capable for calls,” Cao said.

One mode the startup has built its room of voices is by simply asking users to taxable their ain voices for grooming its models, and compensating them if their voices are used. That resulted successful immoderate occupation a fewer months ago, however, arsenic some creators alleged that their voices were uploaded to Fish Audio without their consent. The startup had a DMCA contented take-down process successful spot to code specified concerns, but the take-downs themselves took a agelong time.

Fish Audio’s CEO and co-founder Rissa Cao told TechCrunch that the institution has present automated the take-down process. Creators tin easy taxable a abbreviated dependable illustration oregon a declaration to beryllium that an uploaded dependable belongs to them, and their dependable volition beryllium taken disconnected the startup’s level successful little than 3 minutes, she said.

Still, that doesn’t forestall anyone from uploading an artist’s dependable without their knowledge. And until the creator finds out, their dependable volition proceed to beryllium utilized connected the level until they record for it to beryllium taken down.

Oskue Honda, a spouse astatine Coreline Ventures, said a community-driven exemplary lone works erstwhile creators spot the platform.

“A community-centric attack tin lone go a durable vantage if creators spot the platform. That means consent, transparency, and attribution indispensable beryllium built into the merchandise alternatively than treated arsenic afterthoughts. I judge the manufacture needs to determination toward verified dependable ownership, wide licensing terms, casual reporting and takedown processes, and yet revenue-sharing models wherever creators payment financially erstwhile their voices are licensed oregon utilized commercially,” helium said.

Cao said erstwhile the startup was lone offering its merchandise arsenic an open-source task with plans for creators, it was moving efficiently and didn’t request money. But it wanted to make much precocious models, and besides wanted to accommodate enterprises arsenic capitalist involvement was ramping up, which led it to question capital.

Looking ahead, Fish Audio plans to merchandise an audio knowing exemplary this year. It’s besides gathering a speech-to-speech model.

The code procreation marketplace is crowded, with companies similar ElevenLabs, WellSaid, Cartesia, Speechify, Async (previously Podcastle), and Krisp competing for creators and enterprises’ wallets.

According to Rico Mallozzi, a spouse astatine 359 Capital, fine-grained controls for developers and cost-efficient exemplary grooming volition assistance Fish Audio vie amended with large AI labs.

“I deliberation what they’ve been capable to build, state-of-the-art models, with the squad they have, compared to immoderate of these different well-funded AI labs oregon companies, is incredible. It shows their method acumen successful closing the spread betwixt artificial-sounding and human-like voices,” Mallozzi told TechCrunch implicit a call.