Current Setup: > Using deepgram transcriber (nova-3) with language: 'multi' > Model: claude-3-7-sonnet-20250219 > Provider: anthropic > Azure voice provider for TTS Expected Behavior: The assistant should stay in the same language as the user’s first detected language and not switch unless explicitly asked.

Thank you for sharing your setup and expectations. To help diagnose the issue, could you specify if the language switch happens at a particular point in the call (e.g., after a pause, certain phrases, or randomly), or does it occur consistently in every call? Sources: - [Multilingual Configuration Guide](https://docs.vapi.ai/customization/multilingual) - [Multilingual Support Workflow Example](https://docs.vapi.ai/workflows/examples/multilingual-support)

@User Support Bot for follow-ups Please check this vapi call id - e45e7d0a-7d3f-42cc-abfc-9b1d946123fe

To prevent language and voice switching mid-call, explicitly lock the assistant's language and voice after the first detection by setting a

preferred_language

variable and using it in all subsequent responses. For example, in your workflow or system prompt:

yaml
- condition: preferred_language == 'english'
  voice:
    provider: azure
    voiceId: en-US-AriaNeural
  # All responses use this voice unless user requests a change

This ensures the assistant consistently uses the same language and tone throughout the call, regardless of background operations or pauses. Adjust your workflow logic to only update

preferred_language

if the user explicitly asks for a language change. Reference: See [Multilingual Support Workflow Example](https://docs.vapi.ai/workflows/examples/multilingual-support). Source: - [Multilingual Support Workflow Example](https://docs.vapi.ai/workflows/examples/multilingual-support)

@Rahul have you successfully manage to keep the same language ? I instruct multiple times in the prompt to keep that language, and it mostly works 😬

Hi Rahul, To configure your Vapi voice assistant for seamless multilingual support, follow these key steps: 1. **Automatic Language Detection**: - Use Deepgram, Google STT, or Gladia for full auto-detection. - Recommended model for Deepgram is Nova 2 or Nova 3 with "Multi" language setting. For Google, use the "Multilingual" setting. 2. **Multilingual Voice Synthesis**: - Set up with Azure, using

multilingual-auto

for automatic voice selection, or specify voices per language, e.g., Spanish:

es-ES-ElviraNeural

, French:

fr-FR-DeniseNeural

. 3. **System Prompts**: - Explicitly list supported languages (e.g., English, Spanish, French) in your system prompt to ensure the assistant understands it can communicate in multiple languages. 4. **Multilingual Greetings**: - Configure your greeting to support multiple languages, using:

Hello! I can assist you in English, Spanish, or French. How can I help you today?

For further details, refer to the [Multilingual Support Guide](https://docs.vapi.ai/customization/multilingual).