Providers
eleven_multilingual_v2
The ElevenLabs text-to-speech model eleven_multilingual_v2, with every setting it accepts, the allowed value and the default.
The quality-first multilingual model, and the only ElevenLabs model that honours style exaggeration and speaker boost. Noticeably more latency than Flash or Turbo, so it suits recorded prompts better than live turn-taking.
Reached through the ElevenLabs API with OnCore's own API key, not the LiveKit Inference gateway.
Voice settings are set on the assistant, not in the voice body. Leave every one unset and ElevenLabs' own defaults apply; set any one of them and OnCore sends stability and similarityBoost too, at the same values ElevenLabs defaults to.
Settings#
Every setting is optional. Leave one out and OnCore sends nothing for it, so the Default applies. If invalid is what the provider does with a value it does not accept.
| Setting | Type | Allowed | Default | If invalid |
|---|---|---|---|---|
stability | number | 0 to 1 (0% to 100%) | 0.5 | Refused, call fails |
similarityBoost | number | 0 to 1 (0% to 100%) | 0.75 | Refused, call fails |
speed | number | 0.7 to 1.2 | 1.0 | Refused, call fails |
style | number | 0 to 1 (0% to 100%) | 0.0 | Refused, call fails |
useSpeakerBoost | boolean | true or false | true | Used as sent |
Overriding on a single call#
Return this from your assistant-request webhook to change these settings for one call only. The stored voice is left untouched, and the next call starts from it again.
{
"data": {
"overrides": {
"voice": {
"settings": {
"stability": 0.5,
"similarityBoost": 0.75,
"speed": 1,
"style": 0.3,
"useSpeakerBoost": true
}
}
}
}
}Send only the keys you want to change: anything you leave out keeps the value the assistant already has. Add "id" beside "settings" to swap the whole voice for another one from your catalogue at the same time.
Two of these have a usable range narrower than the accepted one. stability below 0.3 makes delivery erratic between renders, and a very high similarityBoost introduces audible artifacts. Both are accepted at any value in range, so nothing warns you; tune them by ear on a test call.
style is the setting unique to this model, and it has the same shape of trap: above 0.5 delivery becomes erratic and latency rises. useSpeakerBoost raises similarity to the original speaker at some cost in generation speed.
Next steps#
- eleven_v3: the next ElevenLabs model.
- Flash and Turbo: the previous ElevenLabs model.
- Providers introduction: per-call overrides, and how a wrong value behaves across providers.