Rissa Cao, Fish Audio CEO on Building AI Voice Infrastructure

Grace Gong (0:01)

We are live. Hi, Rissa. Welcome to Venture with Grace.

Rissa Cao (0:04)

Hi, Grace. Thanks for having me.

Grace Gong (0:07)

I'm so excited for our conversation because I feel like you guys just won one of our top 100 companies this year for Smart AI 100, for the Most Dangerous Competitor Award. And I'm super excited to dive into our conversation.

Rissa Cao (0:23)

Hi, guys.

Grace Gong (0:24)

Okay. So I want to start to give a shout out to our amazing sponsor. This episode is brought to you by Nebius, the ultimate cloud for AI innovators. Nebius provides AI infrastructure you can count on, combining reliability and speed with flexibility and engineering support unmatched by hyperscalers. AI leaders like Meta, Shopify, and Higgsfield already partner with Nebius to run their AI workloads. Plus, Venture Back Startup can save up to 150K on compute costs when they apply for access. Visit nebius.com or nebius.com/start up to learn more. I want to start with your journey because I feel like the first time we met, it was probably at our summit last April, and I feel like you guys grow super fast lately and you guys already hit $8 million ARR and raised $52 million funding for Seed. Now, I think you guys reach approximately, sorry, it was 8 million users in the first year and the $21 million ARR.

I want to start with such a high growth story. I want to start with the founding story of Fish Audio.

From the other podcasts I was listening to, it sounds like you and your co-founder started as a path project, and then he started working on this project and he convinced you to join this project, and then now you're the CEO. Maybe we could talk about the founding story of Fish Audio for the audience who first time heard about it.

Rissa Cao (1:46)

Yeah, absolutely. I think it's right about a year ago, back in July 2025

That's where Amina and Shijia joined force together. Previously, Shijia was my co-founder, who was also the chief scientist, who was operating Fish as an open-source project.

It was a small team of three all coming from top contributors from the open-source community, and all of us are Vtubing fans.

Initially, when we were running this, he wasn't thinking about running this as a VC-back company at all. He really just was really bored of how voice or sounding so robotics, when he was watching all of the Vtubing or the games that he was playing. So, at the time, I was operating another startup in the enemy AI space. And previously, I was working product at Amazon and Meta for the Reality Lab.

From there, I was working on many of the voice product as well. But never I feel like the voice was natural enough to support the production-grade voice that we were looking for. So, me and him, initially, funny enough, I actually was planning to recruit him to join my last company as Chief Scientist, because I think he's one of the most talented people I've ever met in my life in terms of research. And from there, I ended up being supporting him as advisor and investor for his prior research lab. It was called Fish Speech, and it was pretty popular. Open source had more than 100K Github stars already. And when we started operating Fish last summer as a company, we actually operate as a commercial platform on the consumer side. So it's a voice AI generation platform.

To begin with, the initial user coming from the open source gaming community, coming from many of the Vtubing fans, as well as many of them are video and podcast creators, I would say. And gradually, through different growth engine, growth channels, we break through to 3 million ARR, 5 million ARR, and then coming to 10 And that's when we started operating more to the API side. And many of the API customers coming to us also brought us many of the enterprises. We never knew what is the MSA, what is their data retention or DPA at that time. But because enterprise customers started asking us to sign in contract with them, so we started asking legal to prepare all of the contracts, and then starting selling to enterprise customers. As of today, more than 50 percent of our revenue are actually coming from enterprises. And we grew really quickly over, I would say, for the last eight months on the B2B side. And that became a major driver of the revenue growth.

On the creator and developer side, we're serving more than eight million creators, developers as of today, and they're still growing. And just last month, actually, we were live on OpenRouter and we became the most requested speech model on OpenRouter as well. We took 71 percent of all of the speech model requests on OpenRouter. I think it's really the time for voice. People are adopting voice. Voice is no longer just the output layer. It became the default input layer. Anyone who's building an app, now an AI native app, they make it conversational. Doesn't matter if it's, for example, agents, if it's a gaming character apps or any of the customer support or voice agent platforms. Now everything are in tutoring platforms, are enabling these voice. That's a trend. Not only we're seeing lots of enterprise and developers coming to us from existing voice AI providers, but also there are actually people who are building apps, were only using OM before, but now they're giving their product voice. So that's a trend we are being observing and as we grow, we started to serving more and more of enterprise customers. That made a push our model to have the best latency as well as accuracy than ever. On the consumer and creator side, that's where our initial user started with. For those batch of users actually, they really care about the expressiveness, naturalness and the controllability of the model. So not only the people are using our model, but they are also the people help us to train our model, to do RL to help us model perform better.