Hi team, I’d like to set up my Vapi AI agent to handle both English and French conversations automatically. Goal: When a user speaks in English, the agent should respond in English. When a user speaks in French, the agent should detect that and respond in French. What I need help with: Language detection for incoming user speech.when i enabled multi in deepgram and voice provider as 11 labs along wiht model eleventurbo v2.But there is some issue in efficiency. i need to Ensuring smooth handling of multilingual conversations . Do i need to do api calls for enable language detection. @Sahil @Shubham Bajaj @Vapi Support Bot

According to [Vapi's documentation](https://docs.vapi.ai/customization/multilingual), language detection is automated, so you don't need specific API calls to enable it. Try this configuration for better efficiency:

json
{
  "transcriber": {
    "provider": "deepgram",
    "model": "nova-2",
    "endpointing": 300,
    "codeSwitchingEnabled": true
  }
}

Also, include this instruction in your LLM prompt: "Generate your response in the user's input language only, or default to English if unknown." Sources: - [Vapi Multilingual Support](https://docs.vapi.ai/customization/multilingual) - [Discord Channel Discussion](https://discord.com/channels/1211482211119796234/1330949649720414288)

Message marked as helpful by @! AetherMind! 🎉

Well, Nova 3 is supposed to be able to switch between transcription languages on the go, but its wonky, nova3 is crashing my calls completely on VAPI, 11labs has many voices that support different languages too and I find TTS not to be a problem, I use 2.5 flash. But I don't think transcription/language detection technology is there yet, English and French are quite similar languages as well, which further confuses STT models. You can prompt most LLMs to respond in same language as user is speaking, but I find it work a lot better if I set it in system prompt dynamically once I detect language user speaks in. In my case its cyrillic characters so I just have regex that detects cyrillic symbols, which then sets "You must speak in {language}" in system prompt.

Hey! To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!