Mati Staniszewski
Related people
Episodes
Notes
Cheeky Pint · The world of voice AI, with Mati Staniszewski of ElevenLabs
-
How Did ElevenLabs Make Synthetic Speech Sound Human?
Mati Staniszewski says ElevenLabs voices sound human because the models were trained on how speech is delivered, not only on the words.
Audio is easy to find and usually badly labeled. It often does not say who is speaking, or what emotion and action are in the line. ElevenLabs built its own labeling team, coached on voice, and mixed automatic labels with manual ones. A speech-to-text model began as an internal tool, because models already on the market were not good enough to annotate that audio. A researcher on the team later turned that internal model into a product customers could use. Once the voices were in production, the way customers used them and corrected them became more trainin
-
How Does an Audio Model Turn Text Into Speech?
Mati Staniszewski says a modern audio model predicts the next sound from the sounds and the text around it, rather than copying a human throat.
Early systems tried to rebuild the vocal tract as an analog machine. Bell Labs later turned speech into a structured set of signals. Systems after that stored phonemes, the small pieces of a spoken sound, and stitched on the most likely next one. For streaming speech, today's model predicts the next sound from what came before. When the input is text, it uses that text too. A co-founder brought ideas from transformer models and diffusion models into speech so the result can be reliable, high quality, and fast. Predicting the next token, which langua
-
Why Is Voice AI Harder Than Text?
Mati Staniszewski says voice has not passed a conversational Turing test, even though text models did that a long time ago.
A useful voice agent has to decide when a person has finished, when to wait, when to ask a clarifying question, and when to go do the task. That takes speech-to-text, a judgment of turn-taking from silence and from context, and a way to speak or show text before it acts. He thinks a customer-support call can already pass a voice Turing test inside that narrow job. An interactive game, where the agent should feel like another person, has not. The ordinary failure is orchestration. A simple reply can come back at once, but a database lookup, or a step that has to authent
-
How Does Ukraine's Diia Service Use Voice?
Mati Staniszewski says Ukraine's Diia service became a way to reach citizens during the war, and ElevenLabs added voice so people could use it by speaking.
People in a war zone cannot rely on the same phone lines or the same offices. Ukraine had built Diia, a central government service, over the years and leaned on it harder once the war started. It covers benefits, what is happening on the front line, education, and booking a healthcare appointment. ElevenLabs went to Kyiv and made that service available by voice. The lesson he brought back was the staffing. Every ministry had technical people building an agent version of its own work, and a central digital team assembled those pieces for
-
Why Do High-agency People Win From AI?
Mati Staniszewski says high-agency people are the ones who win from advances in AI, and that inside organizations low-agency people will lose out.
ElevenLabs hires for that. He wants first principles, ownership, excellence, and humility, because people can be scaled and culture is much harder to scale. As the company grows, the result he is most proud of is the people and the culture, not any single person or any single product. Culture is what builds the company now. People try to be excellent at the craft and still enjoy the work. He thinks that mix of agency and enjoyment is the best thing ElevenLabs has done.
-
What Is ElevenLabs Annual Recurring Revenue?
Mati Staniszewski says the most recently announced ElevenLabs revenue figure was 350 million at the end of 2025, and the latest quarter added 100 million in net new annual recurring revenue. John Collison put the figure in the 450 million range, a several-fold increase versus the same time a year earlier.
The best proof the technology is working is a wide set of deployments with Deutsche Telekom and T-Mobile, Revolut, Klarna, Meta, and IBM. Over 50% of the business is now sales-led enterprise. The technology behind agentic interactions became reliable and high quality over the last year to year and a half, so an account starts and then keeps expanding in a land-and-expand motion.
When a cu
-
Why Are ElevenLabs Voice Models Cheaper to Train?
Mati Staniszewski says ElevenLabs voice models are definitely cheaper to train than language models and image and video models because they are significantly smaller. A leading-edge voice model has a few billion to the low tens of billions of parameters. Leading language models run in the hundreds of billions of parameters, with smaller variations used at lower scale.
Capital spending is still high. ElevenLabs recently raised half a billion at an 11 billion valuation to keep building the best models and hire the best researchers. His co-founder leads the audio work. Inference cost follows usage. Text-to-speech is usually charged per text token. A voice agent or transcription is charged per
-
Can People Who Lost Their Voice Speak Again?
Mati Staniszewski says people who lost a voice can get speech back, and he calls that the most important result of the work.
The cases he names are people who lost speech to ALS, people who lost speech to throat cancer, and a recent patient with a Neuralink who spoke to family in their own voice. One woman lost her voice before she got married. When the technology could finally recreate it, she spoke the vows at a marriage ceremony for the first time. He describes that moment as heartfelt, and as the clearest reason the speech work matters.
-
Why Do Sales Leads Share More on a Voice Call?
Mati Staniszewski saw people registering interest with ElevenLabs share more once they could speak with an agent instead of only finishing the written form. Voice changed how they dealt with the business.
At a webinar, people were much more keen to get through the form by speaking with the agents, and the form went a lot easier. They were also more open-ended about the use case, giving a wider set of uses and the complexity of the work. Writing it out had been tedious and tricky. Speaking worked like an open-ended adventure: follow-up questions and clarification were possible, and people were more at ease and could trust the system was working. The same path works across all languages, so l
-
Why Can't an AI Audiobook Sell on Audible?
Mati Staniszewski says an AI audiobook cannot be sold on Audible. Audible blocks AI content, and the other companies they tried would not accept an AI audiobook either.
Authors who could not pay for a professional narrator started showing up at ElevenLabs after the first software shipped in 2023. They wanted a book and had no place to sell it. There was no distribution for an AI audiobook, so ElevenLabs built the avenue and called it 11 Reader. A person uploads a PDF or other text and hears it read aloud. The voices run from a reading in the style of Michael Caine through to other performances.