Eleven v4 adds prompt-based control of emotion, pacing and sound effects across more than 90 languages, while Turbo targets low-latency agents.
ElevenLabs launched Eleven v4 and its low-latency Eleven v4 Turbo variant on September 28. The standard model focuses on expressive text-to-speech, while Turbo is intended for latency-sensitive uses such as conversational agents. Both are available through ElevenAgents, ElevenCreative and ElevenAPI.
The central addition is finer control over delivery through natural-language directions and inline tags. Users can specify tone, pacing, emotion, accents and the treatment of individual phrases, or add cues such as laughter, light rain and a buzzing phone. ElevenLabs says the new architecture follows these instructions more accurately than earlier models and offers significantly improved support for International Phonetic Alphabet phonemes for custom pronunciations.
Eleven v4 also considers the context of an entire scene instead of treating each line as an isolated generation. According to the company, this lets speakers respond to preceding dialogue while preserving the distinct qualities of each voice. ElevenLabs positions the capability for audiobooks, character performances, advertising, dubbing and localized conversational agents.
Both versions support more than 90 languages. ElevenLabs says a voice recorded in one language can speak the others while retaining its identity and adopting a native accent. The company also claims stronger accent adherence over the course of a generation, although its launch article does not provide independent multilingual validation.
For Eleven v4 Turbo, ElevenLabs reports a median time to first audible speech of about 150 milliseconds. Its September 2026 comparison used identical scripts and default settings, WebSocket streaming for Turbo, and removed measured network latency from every system. The article separately cites roughly 100 milliseconds of median inference latency, so the two figures describe different measurements rather than interchangeable end-to-end performance.
Voice cloning has been updated as well. ElevenLabs says Instant Voice Clones can capture a voice from 10 seconds of audio, while Eleven v4 adds support for Professional Voice Clones. The company also claims more reliable speaker identity across separate generations, regenerated lines and stitched long-form content; these results are vendor-reported rather than independently verified in the source.
The practical change is the combination of emotional direction, scene-aware multi-speaker dialogue and a lower-latency option within one model generation. That could reduce the need to assemble conversations from separately directed lines, but production results will still depend on prompts, voices, languages and integration conditions. ElevenLabs’ comparisons and demonstrations should therefore be evaluated on the intended content and network setup before deployment.
Sources
Originally published by AI Feed: ElevenLabs launches Eleven v4 and Turbo for expressive speech.
Top comments (0)