I have tried all default providers in Vapi, somehow works just 2: 1. Deepgram. Language: Multi Model: Nova 2 It works very bad, always misunderstand the language. 2. Google Language: Multilingual It works better but still not enough, also sometimes misunderstand language and has high latency. What other options are available? I would be very grateful for any advise.

Try 11Labs

For 11Labs I can't choose "Language: Multilingual" and have to choose specific language, and in this case it understand just this specified language and can't switch to another. This is the problem .

Following this. + why is gemini with such large latency?

I have the sme proble with spanish language. The numbers are sometimes sponken in English, but they should be pronounced in Spanish. I haven't been able to find a model or service that pronounces them correctly.

Deepgram - 100ms Gemini 2.0 Flash - 800ms

Have you tried using use AssemblyAI

It supports just English 🤷‍♂️ https://cdn.discordapp.com/attachments/1392824688518103060/1394253090395258932/image.png?ex=687622e0&is=6874d160&hm=9e9923d8850078a3855edd07a0a83aeddfb460b99dfdd6c09b7786df45905c21&

Yes, so Google does 8x latency of other "startups". Bit ridiculous. The Google multilingual is much better than competition, ... but the lag is bad. Any tricks or tips, community?

Hi Andrii Bruhal, To configure a multilingual transcriber, the two main providers with automatic language detection in Vapi are Deepgram and Google. Since you experienced issues with these, especially with language misrecognition and latency, consider the following options: 1. Optimize Current Providers: - For Deepgram, ensure you're using the "Nova 3" model if available, as it may offer improved performance over Nova 2. - Test with different audio qualities or inputs to potentially enhance recognition accuracy. - For Google STT, using the latest models should help with the latency and performance. 2. **Custom Transcriber**: - Consider setting up a [custom transcriber](https://docs.vapi.ai/customization/custom-transcriber) if you need more fine-tuned control. This allows you to integrate another provider or enhance the processing pipeline for better results. 3. **Provider-Specific Models**: - Explore Azure and other available providers for better results in specific languages by manually managing language detection instead of relying on automatic detection. Review each option and adjust settings based on your specific needs and testing scenarios to further improve multilingual transcription accuracy.