Language detection
What listens for the language a caller is speaking on an agent with several languages — Sarvam or Soniox — and Sarvam's detection model.
An agent with two or more languages listens for the language the caller is speaking, so it can follow them into another of its languages. Language detection sets what does the listening.
You'll find it with the agent's other switching settings: in Agent → Languages, beside Switches language, and in the flow builder under Voice & languages → Changing language. It appears once the agent has a second language.
The choices
| Setting | What it does | Choices | Default |
|---|---|---|---|
| Listens for the caller's language with | What hears which of the agent's languages the caller is speaking. | Sarvam, Soniox | Sarvam |
| Detection model (Sarvam) | Sarvam's model for hearing the language. Open it under the choice. | Saaras v3, Saaras v4 | Saaras v3 |
Soniox is offered where your platform can run it. Where it can't, Sarvam is the one choice. On a platform that doesn't offer this setting yet, it shows Sarvam, read only.
An agent already set to Soniox on a platform that can no longer run it keeps it, marked Not available on this platform yet, until you choose Sarvam.
How it works with switching
Language detection decides which language the caller is speaking. Switches language decides when that changes the call's language: as soon as the caller speaks one of the agent's languages, or only when they ask. Each language keeps its own voice and speech recognition, set in Languages.