DEV Community

Jaden Vo
Jaden Vo

Posted on Originally published at infinityme.dev

Eleven v4 Explained: What's New in ElevenLabs' Voice Model

#ai #tts #audio #machinelearning

#ad #ElevenAgentsPartner

Disclosure: the full guide contains affiliate links. I'm not affiliated with or employed by ElevenLabs.

On September 28, 2026, @elevenlabsio launched Eleven v4, which it calls its most emotive text-to-speech model yet, alongside a low-latency sibling, Eleven v4 Turbo. This is a short summary based on official ElevenLabs sources (launch post, product page and docs), not a hands-on review.

What's new

  • New architecture: ElevenLabs says v4 reads a script "the way a voice actor would", interpreting tone, pacing, emotion and context.
  • Direction in the script: inline audio tags like [laughs], [whispers] or [door slams], or natural-language delivery notes. ElevenLabs says v4 follows these more reliably than v3.
  • Multi-speaker dialogue: the model understands the whole scene, so speakers respond to each other.
  • Consistent voices: a new method for capturing speaker identity keeps a voice stable across regenerations, dialogue and narration.
  • Long-form: context stitching keeps pacing steady in long scripts such as audiobooks. One request handles up to 10,000 characters (about 10 minutes of audio).
  • 90+ languages.
  • Professional Voice Clones return: not supported in v3, they now work with the full emotional range.
  • Pronunciation control: improved IPA phoneme support plus Pronunciation Dictionaries for names, acronyms and technical terms.

v4 vs v4 Turbo

  • Eleven v4 (eleven_v4): best for expressive or long-form narration, audiobooks and character voiceovers. Available in the ElevenLabs app and API.
  • Eleven v4 Turbo (eleven_v4_turbo): built for real-time experiences and voice agents, with median inference latency around 100 ms and time to first speech around 150 ms, per ElevenLabs. Available in the API and ElevenAgents.

Both support audio tags.

Who it suits

Audiobook narrators, game and animation creators, YouTubers and ad producers, developers building voice agents, and multilingual teams.

How to start

  1. In the ElevenLabs app, open Text to Speech and choose Eleven v4.
  2. Pick a voice, paste your script and add direction with tags, e.g. [warm] Welcome back. [long pause] [whispers] Here's the secret.
  3. Via the API, pass model_id: "eleven_v4" (or "eleven_v4_turbo" for real time) to the Create speech or Stream speech endpoints, or use Create dialogue for multiple speakers. TypeScript and Python SDKs are available.

Full guide: https://infinityme.dev/blog/eleven-v4-elevenlabs-new-voice-model/

Top comments (0)