I set up for an assistant via api. I can see in the code but when i make a call i see on the logs using deepgramm stt not mine and on the dashboard i see error that i have to select a transcriber from the list. But there is no custom model. I used this to patch my old assistant: curl -X PATCH https://api.vapi.ai/assistant/my-assistant-id \ ─╯ -H "Authorization: Bearer my-api-kex" \ -H "Content-Type: application/json" \ -d '{ "transcriber": { "provider": "custom-transcriber", "server": { "url": "wss://mywebsite.com/api/custom-transcriber", "headers": { "x-vapi-secret": "my-api-key" } } } }' and on the dashboar when I open the assistan and check the code I can see it has been set: "transcriber": { "server": { "url": "wss://mywebsite.com/api/custom-transcriber", "headers": { "x-vapi-secret": "my-api-key" } }, "provider": "custom-transcriber" }, CALL ID: 019c89e8-12b6-7112-9ce5-b6d6c2ba1cf0

Thanks for sharing your setup details. To help diagnose the issue, could you please check if your custom transcriber server at

wss://mywebsite.com/api/custom-transcriber

is receiving any connection attempts or logs from Vapi when you make a call? This will help determine if the connection is being established correctly. For reference, your server should log a

"start"

message from Vapi and stream audio data as described in the [custom transcriber documentation](https://docs.vapi.ai/customization/custom-transcriber). Source: - [Custom transcriber documentation](https://docs.vapi.ai/customization/custom-transcriber)

No i did not see any request from vapi

same issue here. the request not sent to custom transcriber server checked the call logs, it use deepgram as transcriber @User please fix it

Hey guys, just an update that the team is working on this issue and it'll be fixed soon.

Ah that's a really great news because I need this ASAP for my client

Hi, Sorry for the delay, and thank you for your patience. Could you please share a recent call ID where you experienced this issue? This will help us investigate it more effectively.

We're running a custom transcriber server connected via your WebSocket transcriber protocol. We're seeing a consistent issue where the start message is never delivered to our server — only raw audio binary frames arrive, starting immediately after the WebSocket connection is established. What we observe: - WebSocket connection is established successfully - Binary audio frames begin arriving ~15ms after connection - No start message (JSON, type: "start") is ever received - The session receives audio for the full call duration with no initialization message What we expect per your docs: - A start message with encoding, sampleRate, channels, etc. should arrive before or shortly after audio begins Impact: - Our server cannot initialize the downstream transcription service without the audio format metadata from start - The entire call goes untranscribed Environment: - Partner: David STT-TEST - Example session with this issue: 03de1449-a301-4332-8e2e-03ac6b40026e (2026-03-24 ~19:20 UTC) Could you confirm whether the start message is expected to arrive before audio, and whether there's a known issue with it not being sent? A minimal example of the correct message ordering from your side would also help.

Hey, Custom TTS cannot be configured through the Vapi dashboard GUI, it is an API-only feature. You can Create Custom Credentials in the Vapi dashboard and Use the Vapi API to create or update an assistant with custom TTS configuration