ElevenLabs launches v4 speech models with support for 90 languages
Creators get finer control over delivery, while voice-agent replies can start before an answer is complete.
ElevenLabs has launched its v4 and v4 Turbo speech models, designed to give users more control over delivery and reduce lag in voice-agent conversations. A ten-second sample is enough to create a voice clone, and creators can layer expression cues in sequence while the system tracks context across longer passages. The models support 90 languages, up from 70, with ElevenLabs reporting its largest quality improvements in Japanese, Brazilian Portuguese, Mandarin and Cantonese. For agents, v4 can begin speaking before the connected LLM has finished its answer; the company also says it handles escalations, confrontations and waiting periods differently. ElevenLabs’ annualized revenue run rate has grown from roughly $330 million at the start of the year to more than $600 million.
Why it matters
For creators, the shift means more say over how a spoken passage is delivered. For people talking with voice agents, audio can begin before the connected model has finished its answer, making the exchange feel less like waiting for a full response. The wider language support also brings the models to more language communities.
Signal or noise?
Does this story matter, or is it hype? Decide before you see what everyone else thinks.
Sources
- TechCrunch