This is really critical as our production users are affected. But initially, the team used dashboard mainly to configure the assistants for initial launch. We decided to switch to api based approach, with the config that 1:1 mirrors the dashboard one. Basically after customer pick ups the call, the assistant drops it after few seconds of interaction and throws

call.in-progress.error-providerfault-eleven-labs-500-server-error.

We haven't explicitly configured to use any of the eleven labs llms. Some of the affected call ids:

38bff7c4-ebd9-4866-a381-d6638108912e
47b709c2-c515-4206-b523-46d509e2a7a8
4ff90d59-19aa-472f-927e-6bf0b2081ca3
2b033a2e-f6ef-494b-bba2-6b178e9ebfa2
155bdda5-25a6-45bb-b7b3-83843c2e99f6
155bdda5-25a6-45bb-b7b3-83843c2e99f6
471383dc-d3b7-416a-bf30-b74ba225d352

https://cdn.discordapp.com/attachments/1385369483312369694/1385369483681730670/Screenshot_2025-06-20_at_01.23.02.png?ex=6855d15f&is=68547fdf&hm=0c086c0f4973f3f712e365a15c8eb2f0fe91dff85839e7ef7308fa58ad4d0741&

It appears that your calls are failing due to a server-side error from the ElevenLabs TTS provider, even though you haven't explicitly configured it. To ensure a specific TTS provider is used, explicitly set the

voice.provider

field in your assistant configuration when making API calls. For example:

json
"voice": {
  "provider": "azure",
  "voiceId": "en-US-JennyNeural"
}

This should prevent Vapi from defaulting to ElevenLabs. If the issue persists, please contact Vapi support with your call IDs for further investigation. [Reference: Call Ended Reasons](https://docs.vapi.ai/calls/call-ended-reason) Sources: - [Call Ended Reasons](https://docs.vapi.ai/calls/call-ended-reason) - [ElevenLabs Integration](https://docs.vapi.ai/providers/voice/elevenlabs)

the config is already explicit with the configuration of voice provider (and its not eleven labs) https://cdn.discordapp.com/attachments/1385369483312369694/1385370495754698864/Screenshot_2025-06-20_at_01.27.06.png?ex=6855d251&is=685480d1&hm=36da67aebd5db3ef65dff13338637d188fc64c18b0039375ff6c0cb9c1f377f2&

Looking into it

I see many people have/continue to report the same issue. Perhaps it should by treated as a high priority incident? @User @Kings_big💫

Are you using Vapi 11labs key?

@Sahil no

Looking into it.

@Shubham Bajaj please, let me know if there's anything I can do from my side to help to resolve this issue quickly

🔵 17:07:17:113 Couldn't Generate 11labs Speech. Error: {"detail":{"status":"unsupported_language","message":"Model 'eleven_turbo_v2_5' does not support the language_code en-US."}} 🔵 17:07:17:113 [user ERROR] Voice failed (reason: {"detail":{"status":"unsupported_language","message":"Model 'eleven_turbo_v2_5' does not support the language_code en-US."}}) 🔵 17:07:17:019 Generating 11labs speech using elevenLabsSpeechStream { "text": "Hey there fellow traveler", "model_id": "eleven_turbo_v2_5", "language_code": "en-US", "voice_settings": { "stability": 1, "similarity_boost": 1, "speed": 1.03 } }

@Shubham Bajaj interesting, the sdk typings have this value baked in, otherwise compiler would have thrown an error. Also this should have been caught earlier in the process, likely by a rejection on POST /call with a proper validation message. This way it just obfuscates the underlying issue for the consumers with that 500 error. Thanks @Shubham Bajaj ! I'll patch up the code, do the testing, and let you know how it went!

6b55e3ec-6448-4679-9731-503526bd5a55

Hey Yuko, can I mark this ticket as solved?

hey @Shubham Bajaj . Yes, tested it, seems to be working fine now, go ahead

Marking this ticket as Solved ✅