Hi Vapi team, I've been using your service for a while now, and I really appreciate the product overall. We're currently testing the system for German language conversations, but unfortunately we're facing a consistent issue: In some of our calls, the voicebot does not seem to hear the customer properly, and the customer has to repeat themselves multiple times, even though the audio is clearly captured and audible from the microphone input. We are not sure whether this is due to latency issues for German, a limitation in the transcription model, or something else entirely — but this issue has prevented us from using the system in production. Here are our setup details: Model: OpenAI GPT-4o mini cluster Transcriber: Deepgram Nova 2 (language: de) Voice: 11labs, eleven_multilingual_v2 Call ID (where the issue is clearly visible): 6bea2649-e632-4ec6-8077-439262a809d1 We would appreciate it if you could take a look at this and help us understand what might be going wrong, or if there are any recommended changes to improve German-language reliability. Thanks a lot in advance!

So, I faced similar issues when I was trying to make it work with Spanish and stuff like that. I think it's not an issue with VAPI—it's just an issue with the model, the transcriber, and the voice you are using. And that's the funny part: you just have to try. You just have to go into your VAPI dashboard, try different model, transcriber, and voice to see what's the best for German. It's just like a play game—this is not really a VAPI issue. For the model, I would suggest using GPT-4-1 mini because it has cheap as 4o and it's also really quick. For the transcriber, you should try Google Flash 2.0. It's really good. (I was impressed)-i think, you just have to say mutlilingual or some stuff like that For the voice, make sure that you go into Eleven Lab and try and find a voice that works with German,? Don't forget that you also have the possibility to add custom voices. You could also go into Eleven Lab, find a really good voice for German, and then add your API key, and then you will have something that sounds even more natural. The Tl;DR is that you have to try out. This is not really a VAPI issue, I think (let me know if you are able to solve it).

I will have to agree with @jojo here, switching providers to even google and azure can make a difference for transcribers and voices, but also changing the LLM might help too with anthropic or even grok. Try some combinations out and see what works. If you still don't get good results, you can always come back here to get any suggestions or answers