Hey! We connected a custom transcriber (Soniox) to Vapi, but the transcriber only hears the user, not the AI so the AI keeps looping because it doesn’t know what it just said. It looks like the same issue as here: https://discord.com/channels/1211482211119796234/1413062535808155759 Could you check if there’s a fix or something we might be missing? Here are some call ids : 0199bfa3-095a-7aa7-aa42-0ea3aa27cf66 0199bf93-33af-733b-9465-267a5141f85f 0199bfca-cd02-7660-afd9-d948e826cb3f

It sounds like your custom transcriber is only processing the user's audio channel. To fix this, ensure your code detects and emits transcripts for both channels. For example, in your transcript event handler, use logic like:

js
const channel = channelIndex === 0 ? "customer" : "assistant";
// Emit transcript for both customer and assistant
emitTranscription(text, channel);

Make sure your transcriber processes both

channel_index: 0

(user) and

channel_index: 1

(assistant) so the AI can "hear" itself and avoid looping. See the [Vapi custom transcriber guide](https://docs.vapi.ai/customization/custom-transcriber) for a full example. Source: - [Custom transcriber integration guide](https://docs.vapi.ai/customization/custom-transcriber)

@User we are kinda on a short deadline help would be much appreciated

cc: @kyle

hey there, we looked at the logs and aren't seeing any transcription data from the user or LLM

{
  "role": "user",
  "time": 1759855725612,
  "endTime": 1759855725612,
  "message": "",
  "duration": 0,
  "metadata": {
    "wordLevelConfidence": [
      {
        "end": 0,
        "word": "",
        "start": 0,
        "confidence": 1,
        "punctuated_word": ""
      }
    ]
  },
  "secondsFromStart": 0.756
}

Yeah we resolved that but we have a problem that Asistnant can hear him self and basiclly he talks to him self you can se that in this call 0199ce51-a6bc-7aa9-864e-afe8008b9606

How did you guys do it at the end? Working on it too. Really need help @ksjakj @kyle @Shubham Bajaj @User

@Blacky we didn’t bro

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.

@kyle tried, but I definetely need in depth support, as soniox seem to be the best transcribing model for europeans

Thanks for the update — based on what you’re describing (the assistant hearing itself), this is almost always caused by insufficient echo-cancellation at the SIP transport layer rather than an issue with Soniox itself. Here are a few things you can try: 1. Strengthen Echo Cancellation in Your SIP Setup Since you’re using SIP, please confirm the following with your SIP provider: - Echo cancellation (EC) is fully enabled - Acoustic echo cancellation (AEC) is supported and turned on at the trunk level - The provider is not mixing the outbound audio back into the inbound channel If your provider cannot guarantee proper echo suppression, this issue will repeat. 2. Try a Different Transport Provider If the echo persists, switching providers can resolve it immediately: - Twilio – built-in echo cancellation, very stable - Vonage – stronger echo handling on international routes - Daily – WebRTC-based transport with excellent echo suppression 3. Optional: Feature Request If you'd like native Soniox support integrated into Vapi (without custom transcriber routing), we can forward a feature request to our team. Let me know if you want us to log that request or if you'd like help testing with another transport provider.