Hey, I am facing an issue with my twilio, I made a vapi assistant for my client integrating with twilio, the assistant works perfectly fine when I test from vapi itself but when I call it from the twilio number it doesn't listen well, doesn't respond as quick or as efficiently as its supposed to, most of the time stops responding right before activating the function call (after receiving the user's info it confirms it then calls the function, here comes the issue, after I confirm it stops responding 90% of the time.) sometimes it also works just fine. Is it an issue with the settings that it doesn't hear me well? Or twilio is just this way? I am using google transcriber (multilingual) and 11labs (Will) voice. @Vapi @Vapi Support Bot

Suggestion: To pinpoint where the issue occurs, go to your Vapi dashboard and review the Call Logs for calls made via Twilio. Look for any errors or delays in the tool execution and check if the transcriber is accurately capturing your input right before the function call. This will help you determine if the problem is with transcription, function triggering, or integration. Tip: Use the "Observe > Call Logs" section to see detailed transcripts and tool results for each call ([Vapi Debugging Docs](https://docs.vapi.ai/debugging#check-logs)). Source: - [Debugging voice agents](https://docs.vapi.ai/debugging)

@User @User @Vapi Support Bot the transciber is not getting the words correctly when I use the twilio number

Thanks for the update. To help narrow down the cause, could you let me know if you notice any difference in audio quality (e.g., background noise, call clarity) when calling from the Twilio number compared to testing in the Vapi dashboard? This will help determine if the issue is related to the audio input or the transcriber settings. Source: - [Debugging voice agents](https://docs.vapi.ai/debugging)

@Vignxt Flow could you help me with that?

Hey! I can help you fix that, I’ve seen this happen often when Twilio’s audio stream or transcription timing desyncs with VAPI. It’s not usually the assistant’s fault but how Twilio handles real-time audio packets. We can adjust buffering, timeout, or transcription settings to make it respond faster and smoother. I’ve worked with 11Labs and Google transcriber setups before, so I know the tweaks needed. are you using Twilio Media Streams or a webhook integration for the voice flow? @Jaafar

Hi Jaafar, For issues with Twilio calls, it may be helpful to verify that your webhook settings are correctly configured in your Twilio account and that your server can receive webhook calls. You might consider checking for any specific error messages associated with Twilio in the [error list](https://docs.vapi.ai/api-reference/calls/get). If you encounter specific errors with Twilio, such as issues with call connections or API key mismatches, refer to the detailed error enums provided in the documentation to identify the cause and resolution. For further integration instructions, you can review the [Twilio phone number DTO related settings](https://docs.vapi.ai/api-reference/webhooks/client-message) to make sure all required fields and configurations are accurately set up. Please also share a call id so we can review the logs

Hey! I am using a simple webhook integration.

If you are only speaking one language, it may help to switch to a different transcriber and set that language. We reviewed the call logs and see that the transcriber interpreted multiple languages. Is there any specific reason you are selecting the multilingual option?

I need to keep it on the multilingual transcriber as this project is for a barbershop and the owner needs a customer support in the romanian and english language. the issue isn't in the transcribers I assume, because when I test from the vapi assistants absolutely nothing goes wrong, issues only occur when I use it from the twilio number.

Thanks for the call id. We will look into it some more and get back to you with some recommendations and observations

Thank you, will be waiting

Thanks for sharing those details. From reviewing the call, it looks like the lower audio quality on phone calls is lowering the transcription accuracy — which explains why it performs better on web calls. In addition, since the multilingual transcriber setting is designed to cover every language, it’s falsely detecting other languages in your transcript. You have a couple of options here: Option 1: Use Deepgram with

language=ro

as primary - Set Romanian as the primary language. - The model can often still understand some English due to cross-language capabilities. - This is better than multi-language for reducing false Telugu/Hindi detections. Option 2: Keep Google Gemini 2.0 Flash but improve prompt engineering - Your assistant already handles this pretty well (it asked for clarification). - Add more explicit instructions to reject non-Latin script characters. - Encourage it to ask for repetition when transcription seems unclear. My recommendation: Since your assistant already recovered well from the error in this call, Option 2 may be the best fit — keep the current multilingual setup but enhance the assistant’s prompt to be more aggressive about rejecting non-Latin characters and asking for repetition.