Description: We are experiencing an issue where partial transcripts generated by Deepgram trigger the endpointing rules and are forwarded to our custom-LLM, causing incorrect messages to be sent before the final transcript arrives. For example (call id below): Partial transcript at Sep 08, 12:17:16.183 Message to custom-LLM (incorrect transcript) at Sep 08, 12:17:17.095 Final transcript at Sep 08, 12:17:17.129 Message to custom-LLM (correct transcript) at Sep 08, 12:17:17.632 This behavior creates multiple issues in our workflow, as only the final transcript is accurate enough for our use case. Request: We would like guidance on how to configure Deepgram so that endpointing and downstream triggers only occur on final transcripts. Ideally, partial transcripts should not trigger endpointing events or be sent downstream. Call ID: d3fe3c30-67fd-4ad8-bb22-8672fc7219e9

I’ve run into this kind of issue before with Deepgram’s partial vs. final transcript handling. The key is making sure your config only triggers endpointing on finalized transcripts — otherwise, partials slip through and cause exactly the duplicate/incorrect behavior you described. I can help you set up the right Deepgram settings (e.g. endpointing, utterance_end handling, and streaming options) so only the final transcript is forwarded downstream to your custom-LLM. Quick question: are you currently using interim_results in your Deepgram stream config, or relying purely on default endpointing? @David Jurado

thank you @Vignxt Flow for your response. We’re currently building our own custom transcriber with Deepgram, since relying only on Vapi’s endpointing rules hasn’t fully solved the issue. For the custom transcriber setup, which configuration or parameters would you recommend using?

Since you’re building a custom transcriber with Deepgram, I can definitely help optimize it so only final transcripts trigger downstream events. Key areas we can look at include interim_results, punctuate, utterance_end handling, and configuring endpointing parameters to avoid partial transcripts from firing. I can assist you on the exact setup to ensure your custom-LLM only receives the final, accurate transcript. @David Jurado

Ok, let's do it. How can you help me?

Kindly let's discuss this more privately @David Jurado

Let us know if you need any additional support from us

Yes, I need support from you, not from an external VAPI member who wants to provide me service through WhatsApp.

Alright then Good luck

We apologize for the delay in responses lately. We are in process of making some changes to our workflow in how we handle support requests. The good news is that we are starting to get back on track of having a 48 hour response time and we would like to start cleaning up anything old. With that being said, just wanted to check in with you to see if you have made any progress or changes to your project since support has last reached out.