Language switches without interrupting the caller
October 5, 2026 · release
When the runner detects a new language, the caller is usually still talking. Until now a wait message could cut in right then. By default it no longer does: the runner holds the caller's speech, starts the new speech services in the background, and speaks only after the caller has stopped.
What the caller hears depends on how the switch went:
- The new language was not ready. A wait message in the current voice, then the answer in the new language.
- It was ready, but the switch took a while. A short ready message in the new voice ("Okay, machen wir auf Deutsch weiter."), then the answer. The threshold is 2 seconds by default.
- It was ready and fast. The answer, nothing else.
If you want the old behavior, set the wait message to interrupt the caller. The dashboard labels that option "Interrupt the caller to announce the switch".
Automatic switching now needs VoiceThere for both speech-to-text and text-to-speech. Only the VoiceThere catalog ties a language to a speech model and a voice, so the runner can pick the pair itself. With a cloud vendor on either side the toggle is disabled in the dashboard and the API rejects the setting. Cloud vendors still work when your agent starts the switch, using per-language profiles.
Every decision of the automatic flow also shows up as a language_switch.* line in the agent logs, and the docs now include a diagram of the whole flow.
Try it
- Docs: Switching language mid-conversation has the flow diagram, the requirements and every log line. System log events lists the fields.
- Dashboard: Project → Voice → Advanced voice pipeline → Spoken language identification (deploy to apply)
- CLI:
voicethere projects voice-advanced set languageId.autoSwitch.readyMessage.minSwitchMs 3000
voicethere projects voice-advanced set languageId.autoSwitch.readyMessage.texts '{"de":"Gut, dann auf Deutsch."}'