Tune spoken language identification from the dashboard

September 10, 2026 · platform

Voice and Voice+Data projects can now tune spoken language identification from Advanced voice pipeline — no runner env hacks required.

What changed

• Enable/disable — languageId.enabled turns LID off while keeping the Whisper model on cloud deploys.
• Minimum speech window — languageId.minSpeechMs (1000–5000 ms, cloud default 2500) controls how much inbound audio Whisper tiny sees before the first user_language event each utterance.

Lower values identify sooner; higher values trade latency for accuracy on short utterances. The SDK and self-hosted runners still default to 1000 ms when unset.

Try it

• Docs: Spoken language identification · Advanced voice settings
• Dashboard: Project overview → Voice (STT / TTS) → Advanced voice pipeline → Spoken language identification
• CLI:

voicethere projects voice-advanced set languageId.minSpeechMs 3000
voicethere projects voice-advanced set languageId.enabled false

Changes apply on the next Deploy to cloud.

Similar posts

• Spoken language identification for voice agents
• Inbound RNNoise on by default for voice runners
• Live showcase demos on the marketing site