We are getting error: Pipeline error deepgram transcriber failed - for all our outbound answered calls. Assistant 1: The transcriber model is set to nova-2, Spanish language. Assistant id: c724b870-422c-4b0a-a9f1-8cef38ba44ff Assistant 2: nova-3, European Portuguese language. Assistant id: f8c3585e-3a3b-42ab-a6d9-342d0fb27faa { "id": "log-10", "level": 50, "time": 1776103097122, "timestamp": 1776103097122000000, "severityText": "ERROR", "severityNumber": 17, "body": "Deepgram WebSocket error", "attributes": { "category": "transcriber", "callId": "019d87fe-6f2e-7001-904a-f2edc6f16c66", "orgId": "27beeeab-8f09-4ea5-ab0f-84916e902325", "provider": "deepgram", "error": { "message": "Deepgram connection failed with status 402 (Request ID: 019d87fe-82f4-7213-af68-e497d83c1869)" } } } Some of the call ids: 019d8d27-ab8c-7000-a28b-1be70d1b5535 019d8d50-3c45-7000-a318-05cc4aacbbfa 019d8d27-ab8c-7000-a28b-1be70d1b5535 Please let us know how to fix it and what's the issue! https://cdn.discordapp.com/attachments/1493710037514977371/1493710038026424452/image.png?ex=69dff560&is=69dea3e0&hm=9430beb14c31acce052d9c941f95dd23959c0a9f348c1e2201855ae7061c893e&

This looks like a billing/usage issue on Deepgram status 402 usually means the transcriber request is getting rejected before it even processes audio, not a config problem. I’d first verify your Deepgram balance/limits, then double-check model + language pairing (nova-2 ES / nova-3 PT) and that your API key has access to both. Also worth testing a direct Deepgram WS call outside VAPI to isolate if it’s provider-side vs pipeline. I can help you debug this end-to-end and pinpoint exactly where it’s breaking are all calls failing or only the ones that get answered? @Batyr

We're investigating this and will get back to you.

I am getting this too

could you please share a call id?

019d9083-ddd6-7bb1-8ff8-deca2213c32b @Shaunak

looking into it

I have switched to a different model for now for transcription instead of deepgram, and that works. @Batyr

yeah, while we look into this, a quick workaround would be to switch to a different model