From the logs I can see: All digits are captured correctly (9 5 1 3 5 1 8 8). The # terminator is also captured. The assistant calls dtmf_tool and finishes successfully (Tool calls finished). The model generates a response (phone number confirmation). ElevenLabs WebSocket opens, BOS (beginning of speech) is sent. But then: ElevenLabs input timeout exceeded → WebSocket closes before any speech is played. So the assistant does generate the text confirmation, but the audio channel closes prematurely and the user hears nothing.

Hi Preben, To set up a custom ElevenLabs voice in Vapi, you'll need to follow these steps: 1. **API Subscription**: Subscribe to an appropriate API plan on the [ElevenLabs pricing page](https://elevenlabs.io/pricing). 2. **Get API Key**: Obtain your API key from the 'Profile + Keys' section on ElevenLabs. 3. **Integrate with Vapi**: - Navigate to the [Vapi Provider Key section](https://dashboard.vapi.ai/keys). - Enter your ElevenLabs API key under the ElevenLabs section and click save. Once saved, your voice library will sync automatically. You can then search for your cloned voice in the "voices" tab in the assistants page, and use it with your assistant. For detailed configuration, refer to [this guide](https://docs.vapi.ai/customization/custom-voices/elevenlabs). Thanks so much for the update and we really appreciate the time you spent investigating this. You’re correct that this issue appears to be related to Norwegian voices on ElevenLabs and other models, and not something directly within Vapi’s control. That said, we’ll make sure to pass your findings along to our team so we can stay aligned with any future improvements or compatibility updates from ElevenLabs. We truly appreciate your persistence in digging into this and sharing what you found it’s super helpful for us and for other users who might run into similar issues.