Article complete

ElevenLabs CEO Says AI Speech May Pass Turing Test This Year

The key trade-off lies between expressiveness and reliability, according to CEO Mati Staniszewski.

ElevenLabs CEO Says AI Speech May Pass Turing Test This Year

Key takeaways

  • ElevenLabs aims to pass the Turing test for AI speech by the end of this year or early 2026.
  • CEO Mati Staniszewski highlights a trade-off between expressiveness and reliability in AI speech models.
  • The company plans to transition to a unified duplex model for more responsive AI interactions.
  • Latency remains a challenge for integrating audio with large language models at production scale.
  • Staniszewski envisions voice as a primary interface for future technology interactions.

Takeaways drafted with AI from the article text.

“We would love to prove that it’s possible this year,” Staniszewski said in a recent interview. “You can cross the Turing test of speaking with an agent, and you just would say this is like speaking with another human.”

The company is currently using a cascading architecture, separating speech-to-text, language generation, and text-to-speech, but is preparing to shift to a unified duplex model. “Soon, the one we’ll deploy will be a truly duplex model,” he said.

According to Staniszewski, the key trade-off lies between expressiveness and reliability. “The true duplex model will always be quicker, a little bit more expressive, but less reliable,” he said. “The cascaded model is definitely more reliable...but maybe not as contextually responsive.”

Latency, too, remains a challenge. “I think we can get pretty good latency on both sides,” Staniszewski noted, while acknowledging that integrating audio with large language models at production scale has yet to be solved. “No company has been able to do it well...I hope we’ll be the first one.”

Citing ongoing work by Meta and OpenAI in the same space, he said, “I don't think it passed the Turing test yet.”

The CEO also reaffirmed his belief that voice will become a primary interface for interacting with technology, signalling a shift in how users may engage with software in the near future.

The company recently launched the alpha version of its new flagship text-to-speech model, Eleven v3, which the company claims is its most expressive model to date. The release introduces inline audio controls, dialogue generation, and support for over 70 languages, targeting creators in film, gaming, audiobooks, and accessibility.

Published

Corrections: AIM corrects errors of fact and notes the change on the story. Spotted one? Tell the editors. Read our editorial standards.

More in this story

elevenlabs · 5 stories
  1. You are here: ElevenLabs CEO Says AI Speech May Pass Turing Test This Year
  2. ElevenLabs Deploys Multilingual AI Voice Support for Meesho for Handling 60,000+ Calls Daily
  3. ElevenLabs Unveils Chat Mode to Expand Conversational AI Beyond Voice
  4. ElevenLabs Appoints Karthik Rajaram as India Country Head
  5. ElevenLabs Doubles Valuation to $22 Bn as Enterprise AI Agent Demand Grows
About the author

Tech Journalist

Siddharth is a media graduate who loves to explore tech through journalism and putting forward ideas worth pondering about in the era of artificial intelligence.

All stories
Belamy · Daily

The day’s AI news, before your first meeting.