Providers
Ink 2 and Ink Whisper
Cartesia speech-to-text: Ink 2 and Ink Whisper, with every setting it accepts, the allowed value and the default.
Cartesia's ink-2 transcriber.
ink-2 decides the end of a turn on its own semantic judgement rather than a silence timer, which makes turn-taking feel quicker. It is English-only here; Ink 2 Multilingual is the same model for French, Hindi, Japanese and Spanish callers. ink-whisper is silence-based, covers around a hundred languages, and is the one to use for any other language.
Reached through the LiveKit Inference gateway.
Both Ink transcribers take the same two settings, which is why they share a page.
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 |
|---|---|---|---|---|
minVolume | float | 0 to 1 | Provider's own | Refused, call fails |
maxSilenceDurationSecs | float (seconds) | 0 or greater | Provider's own | 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 transcriber is left untouched, and the next call starts from it again.
{
"data": {
"overrides": {
"transcriber": {
"settings": {
"minVolume": 0.2,
"maxSilenceDurationSecs": 1
}
}
}
}
}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 transcriber for another one from your catalogue at the same time.
minVolume: Any value is accepted, because ink-2 decides the end of a turn on semantic judgement rather than on volume.
maxSilenceDurationSecs: Not range-checked on either model: a negative or implausible value is accepted and applied rather than refused.
These two settings only take effect on ink-whisper. Both models accept them, but only ink-whisper reads them: it refuses a minVolume outside 0 to 1, while ink-2 accepts 99 and -1 without complaint, which is what a model ignoring the setting looks like. Cartesia's own API reference attributes both parameters to ink-whisper alone, and measuring agrees.
That follows from how each model ends a turn. ink-whisper is silence-based, so a volume floor and a silence ceiling are what it needs. ink-2 judges the end of a turn itself and has nothing to tune from here.
Cartesia publishes no default for either setting. Leave one unset and Cartesia's own internal behaviour applies, with no documented value to state.
The two models also differ in language: ink-2 here is English-only, and ink-whisper covers around a hundred languages. A non-English language on ink-2 does not switch to ink-whisper: the session is refused, and the call transcribes nothing. For callers speaking French, Hindi, Japanese or Spanish, use Ink 2 Multilingual; for any other language, ink-whisper.
minDelay is not a Cartesia setting, so it has no effect on this model.
Next steps#
- Ink 2 Multilingual: the next Cartesia model.
- Providers introduction: per-call overrides, and how a wrong value behaves across providers.