Mati Staniszewski

Mati Staniszewski is co-founder and CEO of ElevenLabs, the voice AI company known for expressive text-to-speech, dubbing, and conversational agents. He met co-founder Piotr Dabkowski in high school in Warsaw, Poland, hacked on side projects while working at Palantir and Google, and turned a shared annoyance with monotonous Polish movie dubs into a research-and-product company focused only on audio.

Meeting Piotr Dabkowski in Warsaw, Poland

They met in an International Baccalaureate class in high school, bonded over mathematics, and stayed friends through living, studying, and traveling together. While Piotr was at Google and Mati at Palantir they ran weekend hack projects: recommenders, crypto risk tools, then an audio coach that analyzed how you speak. On Sequoia's Training Data podcast he traced the founding spark to late 2021: Piotr tried to watch a film with a girlfriend who did not speak English, switched to Polish, and hit the childhood experience of every character voiced by one flat narrator. "It's a horrible experience," Mati said. "We think this will change."

Betting Everything on Audio

ElevenLabs launched into a world that expected foundation-model labs to crush vertical voice startups. Staniszewski's answer on Training Data was focus: "staying focused and staying focused in our case on audio." Little research attention had gone to speech compared with text and images; applying transformers and diffusion to audio, plus building product around the models, is how they competed. Data, he argued, is the hard part: models need transcripts of what was said and also labels for how it was said, including emotion and nonverbal sounds.

Betting on Voice as the Next Interface

On the same podcast he stated the long bet plainly: "voice will fundamentally be the interface for interacting with technology." It carries emotion, intonation, and imperfection that text does not. That thesis pushed ElevenLabs from narration and dubbing into agents for healthcare outreach, customer support, education, and consumer experiments. Quality, latency, and reliability are the three things he says customers buy.

Chasing Human-Level Conversational Voice

He has set an ambitious bar for the research: prove human-level conversational voice is possible soon, whether through cascading speech-to-text / LLM / text-to-speech stacks or true duplex speech-to-speech models. Prosumer releases come first so unexpected use cases surface before enterprise packaging.