ElevenLabs Eleven v4

Use ElevenLabs Eleven v4 and Eleven v4 Turbo as the voice of a pipeline assistant: more expressive speech, audio tags and more than 90 languages.

Eleven v4 is ElevenLabs' newest text-to-speech generation. It reads the tone of what it is given and delivers it: a reassuring sentence sounds reassuring, a question sounds like a question, and it keeps the character of the voice while doing so. It speaks more than 90 languages, Dutch included, and follows audio tags such as [whispering] or [laughing] that your assistant writes into its replies.

It comes in two variants:

Model	Built for
`eleven_v4_turbo`	Conversations. The low-latency variant, made for voice agents. The one to use on a phone line.
`eleven_v4`	The highest quality. Takes noticeably longer before it starts speaking, so it suits assistants where expressiveness matters more than response time.

Both are text-to-speech models, so they apply to assistants that run the pipeline: a speech-to-text model, a text model and a voice. A speech-to-speech provider such as google_realtime, openai_live or xai_realtime speaks by itself and does not use tts_config.

Setup

Eleven v4 runs on VoiceDock's platform keys by default, so you can select it without adding an API key first.

1. (Optional) Add your own ElevenLabs API key

Only needed to use your own ElevenLabs account, for example for voices you cloned there. It is the same key as for the other ElevenLabs models. Add it under Integrations → API Keys.

2. Choose the model

In the dashboard, open the assistant, go to the voice settings, pick ElevenLabs and choose Eleven v4 Turbo or Eleven v4 as the model. Through the API, set it in tts_config:

{
  "tts_config": {
    "provider": "elevenlabs",
    "model": "eleven_v4_turbo",
    "voice_id": "21m00Tcm4TlvDq8ikWAM",
    "language": "nl",
    "stability": 0.5
  }
}

Field	Values	Notes
`model`	`eleven_v4_turbo`, `eleven_v4`
`voice_id`	any voice id from your ElevenLabs voice library	Required.
`language`	language code, e.g. `nl` or `en`	Optional. Omit it and the language is inferred from the text.
`stability`	`0` to `1`	Optional. Lower values give a more varied, expressive delivery; higher values a steadier one.
`apply_text_normalization`	`auto`, `on`, `off`	Optional. How numbers, dates and abbreviations are spelled out before speaking.

Settings per model

The v4 models, like eleven_v3, take stability and apply_text_normalization. The other ElevenLabs fields (similarity_boost, style, speed, use_speaker_boost and apply_language_text_normalization) are used by eleven_flash_v2_5, eleven_turbo_v2_5 and eleven_multilingual_v2, and not by these three models. The dashboard only shows the settings that apply to the model you pick, and saving an assistant with one of the other fields on eleven_v4_turbo, eleven_v4 or eleven_v3 returns a 400 that names the field.

To steer the delivery on v4, use stability and audio tags.

Audio tags

Audio tags are short directions in square brackets that tell the voice how to say the next part. They are not spoken. For example:

[reassuring] That sounds unpleasant, but we can have someone look at it this week.
[whispering] Let me check that for you.
[laughing] That happens to all of us.

The text model writes the tags, so ask for them in the system prompt. Name the tags you want and when to use them, and keep the list short: an assistant that tags every sentence sounds theatrical.

You may start a sentence with one of these tags when it fits the moment:
[reassuring] when the caller has a problem, [cheerfully] when you greet or say goodbye,
[thoughtfully] when you check something. Never use more than one tag per reply.

Only add these instructions to an assistant that runs eleven_v4_turbo, eleven_v4 or eleven_v3: the other ElevenLabs models do not interpret audio tags.

Voices

Every voice in your ElevenLabs voice library works with v4, including voices you cloned. ElevenLabs recommends retraining a Professional Voice Clone that was created before 28 September 2026 to get the most out of v4.

Pricing

Platform keys (default): ElevenLabs characters at cost + €0,07/min orchestration. The rate per model is on Provider pricing.
Bring Your Own Key (optional): €0,07/min orchestration. The characters are billed to your own ElevenLabs account.