The Future of Voice AI: Agents, Dubbing, and Real-Time Translation with ElevenLabs Co-Founder Mati Staniszewski
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
ElevenLabs co-founder and CEO Mati Staniszewski joins Sarah Guo to discuss how his company is revolutionizing human-computer interaction through voice technology. In just three short years, ElevenLabs has grown to $300M in annual recurring revenue by building foundational audio models and products that span creative content and interactive agents. (00:46)
- Core Theme: Voice is emerging as the ultimate interface for everything from content creation to customer service to education, with ElevenLabs leading the charge in making human-AI interaction seamless and natural through high-quality, controllable voice synthesis and interactive agent platforms.
Speakers
Mati Staniszewski
Co-founder and CEO of ElevenLabs, a voice AI company that has achieved $300M in annual recurring revenue in just three years. Originally from Poland, Mati previously worked at Palantir before founding ElevenLabs with his co-founder of 15 years to solve the problem of poor voice dubbing and create more natural human-technology interaction.
Sarah Guo
Host of No Priors podcast and general partner at Conviction Partners, focusing on AI and emerging technology investments. Sarah brings deep expertise in evaluating AI companies and understanding market dynamics in the rapidly evolving voice and agent technology space.
Key Takeaways
Voice Quality Requires Specialized Research Focus
Building truly human-sounding voice AI requires dedicated foundational research rather than relying on existing models. When ElevenLabs started, existing models produced robotic speech that people didn't want to listen to. (07:51) Mati's co-founder assembled a world-class research team to create new approaches from scratch. The key insight is that audio requires different architectural breakthroughs rather than just scale - there are only 50-100 researchers globally who can do this work at the highest level, and ElevenLabs has assembled about 10 of them.
Controllability Matters More Than Out-of-Box Quality
While open source and commercial models are achieving similar baseline quality for content narration, the real differentiator lies in making voice models controllable. (31:19) This means being able to adjust tone, emotion, pacing, and style based on specific use cases and brand requirements. ElevenLabs employs "voice sommeliers" who work with enterprises to select the right voice characteristics for their specific customer base and use cases.
Voice Enables Proactive vs Reactive Customer Experiences
The most transformative customer service applications shift from reactive support to proactive assistance throughout the entire customer journey. (18:37) For example, ElevenLabs works with Micho, India's largest e-commerce platform, where voice agents help customers navigate product discovery, provide personalized recommendations, and guide them through checkout - not just handle refunds and tracking inquiries.
Education Will Be Transformed by Personal AI Tutors
Voice-powered AI tutoring represents one of the most significant upcoming applications, allowing learners to have personalized teachers available on-demand through headphones or speakers. (40:57) Examples include learning chess from Magnus Carlsen or practicing FBI negotiation techniques with Chris Voss. This creates opportunities for celebrity instructors to scale their expertise while providing learners with immediate, interactive feedback.
Research Advantages Are Temporary - Build Ecosystem Value
Technical research provides only a 6-12 month head start before competitors catch up. (32:37) The sustainable competitive advantage comes from building an ecosystem around the technology: brand recognition, voice talent marketplace, integrations with legacy systems, and refined workflows. Companies should use research advantages to accelerate product development and customer acquisition, not as permanent moats.
Statistics & Facts
- ElevenLabs has reached $300M in annual recurring revenue, split roughly 50/50 between self-serve creative subscriptions and enterprise agent platform sales. (02:15) The company serves over 5 million monthly active users on the creative side and several thousand enterprise customers ranging from Fortune 500s to fast-growing AI startups.
- The company has grown to 350 people globally across hubs in London, New York, Warsaw, San Francisco, Tokyo, and Brazil, maintaining a remote-first culture. (02:00)
- ElevenLabs' latest speech-to-text model "Scryvy Two" achieves 93.5% accuracy across the top 30 languages with under 150 milliseconds latency. (35:24) This performance beats existing models on standard benchmarks while supporting nearly 100 languages total.
