00:14:35 · David Senra, Mati Staniszewski
[00:14:35] S0: And you are solely focused on audio? [00:14:37] S1: On the research side, solely focused on audio product. We combine the best of audio with integrations, knowledge, LLMs to help deliver for... [00:14:46] S0: Explain why you think it's so important to be solely focused on audio on the research side. [00:14:51] S1: We think we have incredible talent to be able to go after that. The focus is so important in effectively building the best architecture for those audio models. We think audio is a good combination of not only science but also the arts that you need to solve. It is a little bit subjective. The voices that you produce will be subjective. So you really need to get it right. But ultimately, as you think about AI models, there is data, compute architecture that you need. On audio, we think a lot of the problems that still exist are on the architecture side. We want to be solely focused on solving those architecture problems so you can actually get the best quality out there. A lot of the team, lot of the people that work on our side are the best audio researchers in the world. They're also excited by the premise of being able to continue deploying the best frontier audio models. [00:15:42] S0: And you feel you have technology on the other side that no one else in the world has? [00:15:46] S1: We think so. Yes.
00:15:46 · Mati Staniszewski, David Senra
[00:15:46] S1: We think so. Yes. [00:15:47] S0: It's a weird analogy. I'm so glad I'm reading this book at the same time we're talking because, like, for Honda, they kept kept trying to get him to diversify. He's got other products. Obviously, he builds a bunch of things with engines. But he's like, does that product have an engine? And they're like, no. He's like, then I'm not building that product. He's like, I just do... I focus on engines. And sometimes it's on two wheels and four wheels and everything else, but it's like just engines. It's very similar to what you're saying about audio. And he also said that he felt he had the best engine technology in the world and no one else, could replicate what they did. [00:16:15] S1: There is a common question as we think internally of like, it's 100% true of how you describe it now too. We think about new products, the big question is, do we think we have a unique advantage by applying our audio models in that product experience? If the product experience doesn't have a big bottleneck in that audio and the voice communication side, then it's not our fault. [00:16:39] S0: Explain a situation where you realized that you shouldn't go after that opportunity. Could you give me an example of what you're describing?
00:16:39 · David Senra, Mati Staniszewski
[00:16:39] S0: Explain a situation where you realized that you shouldn't go after that opportunity. Could you give me an example of what you're describing? [00:16:45] S1: The example would... That I'll give you to slightly different spaces. One is on... For for a long time, and I think it's still true, the ideal version in the audio space is if you create a marketing campaign, it's how that combines with a lot of the image and video work to deliver better content. Two years ago, we would have first tried to see whether we can help people create effectively lip dubbing or avatars for for their content. So let's say you switch from one image to another, you need to move the lips, or let's say you want to narrate something, you create an avatar. That was roughly two years ago. A lot of the models that existed on that site just weren't good enough. And deploying a product in this space would be such a defocus and such a shift for the company that it didn't have the audio as a superpower. It still had a bottleneck of the quality of Avatar's quality of a lot of their Because the [00:17:42] S0: problem to solve there is video, not audio.
00:17:42 · David Senra, Mati Staniszewski
[00:17:42] S0: problem to solve there is video, not audio. [00:17:44] S1: Exactly. So this was still poor. So even if you had the best audio applied into that work, still the experience wouldn't be very good. So we decided to not effectively pause any of that effort. Fast forward today is of course shifting now. I think there is a good set of open source models now that exist in that space where the combination of audio, image, and video can actually deliver that result. We are revisiting that today and the recent London ads engine of bringing an ad and bringing that internationally and shifting things in the video, perfect use case of that. But still, a conscious decision we are taking is not creating the model from scratch. So like true text to video or like, you know, VR free type models that exist. We want to be at that intersection where we know the audio can give you the unique advantage, which is usually you already have an existing asset, you need to modify that asset, add that audio component and bring it bring it home. [00:18:36] S0: Do you feel this, like, intense focus or relentless focus on audio is kind of like defense against, the larger labs? For sure. You talk about this in the company?
00:18:36 · David Senra, Mati Staniszewski
[00:18:36] S0: Do you feel this, like, intense focus or relentless focus on audio is kind of like defense against, the larger labs? For sure. You talk about this in the company? [00:18:47] S1: We do. The big part is our focus is audio on the research side and that's where we want to win. I think that we also talk about this in the long, long term, that advantage that we have today will hopefully still be there, but it might not be as big on just the pure research. That's why the other components, that's why the product is so important. That's why the ecosystem is so important that we build around that product. Maybe just to explain what I mean by the ecosystem, of course, there's the brand and trust that you built, but there are also other parts that you can build. In our case, this was investing into effectively a marketplace model where people can create an asset, we authenticate it, and then you can share it and earn compensation as a result. Like trying to create a completely different model for how that works. Give me details about that [00:19:33] S0: because I don't know. Don't know. Yeah.
00:19:33 · David Senra, Mati Staniszewski
[00:19:33] S0: because I don't know. Don't know. Yeah. [00:19:35] S1: I don't know. An example is Voices. Okay. So we did it with Voices where people can create their voice. We authenticate it. Then you can share your voice, your AI voice, and passively as your voice is being used, you earn compensation. We have 20,000 voices today on the marketplace and new people that come in now have a selection of different accents, different styles, different languages, of course, age, gender that you can pick. Every time you use it, other person gets [00:20:04] S0: Do you, used as a company, make these voices, any of these voices, or is it just third parties? [00:20:08] S1: We of course created a system for people to do that. And some of them, when we see pockets that are missing, we'll ourselves try to go after and find people to fill those pockets that are missing and create those voices for the wider ecosystem to benefit. [00:20:22] S0: No one knows about Eleven Reader. You have to do a better job. Was just into your episode with John Carlson, who I love. I don't think he even knew. He kept saying, he's like, Why doesn't this exist? There's an app. Then the follow-up he still talks to us. Like, no, dude. I use that app. What you're asking for, John, is... He already has it. You need to, like, show it to him on your phone while you're recording the podcast with him. So I use George. That's the voice I... In fact, we do all these, like, research reports for the people that come on. [00:20:48] S1: Being a user. That's amazing.