ElevenLabs, the AI voice company, released two new models this week: Eleven v4 and a faster version called Eleven v4 Turbo. The company's pitch is simple. Earlier voice generators could read text aloud, but they sounded like they were reading. These new versions are built to sound like they understand what the words mean and how they should feel, whether that is a tender bedtime line or a panicked shout.
The upgrade most business owners should pay attention to is speed. Eleven v4 Turbo can start producing sound roughly one tenth of a second after it gets a request, faster than the average pause between two people talking. That gap is the difference between a phone bot that feels like a real conversation and one that feels like talking to a machine with a lag.
The models can also clone a voice using only 10 seconds of recorded audio, and then make that voice speak fluently in more than 90 languages while keeping its accent and character intact. A company that hires one voice actor for an English ad can now get that same voice sounding native in Spanish, Japanese, or Arabic, without booking another studio session.
This launch lands while ElevenLabs itself is growing at a pace few startups match. The company closed last year with over 330 million dollars in yearly revenue and was valued at 11 billion dollars after a 500 million dollar funding round in February, more than three times its valuation from a year earlier. Investors including Sequoia and Andreessen Horowitz are wagering that talking to software will become as normal as typing into it, and reports suggest ElevenLabs is preparing for an eventual stock listing.
For most businesses, the obvious use case is customer support. Gartner has estimated that conversational AI could cut contact center labor costs by 80 billion dollars in 2026 alone. A more expressive, faster voice model is exactly what companies need if they want callers to stop hitting zero to reach a human the moment the bot sounds robotic.
But the same technology that makes a support bot sound more human also makes it easier to fake a real one. Cloning a convincing copy of someone's voice from a 10 second clip, shorter than most voicemail greetings, lowers the skill needed to run a scam. The FBI's latest crime report logged more than 893 million dollars in losses from AI related fraud in 2025, much of it tied to voice cloning calls impersonating relatives, executives, and bank staff. Any business that still treats a phone call as proof of who someone is should stop. A phone call can now be faked convincingly, cheaply, and in seconds.
There is a quieter cost too. Voice actors and audiobook narrators, whose entire job is delivering exactly this kind of emotional nuance, are already losing work to AI narration as publishers chase lower costs and faster turnaround. This launch does not create that shift, but it does make it easier for a studio to justify making it.