AI agent voice is choppy with long silence gaps on all real outbound calls — started a few weeks ago, browser test calls work fine

Support Help
1

Every outbound call triggered from our backend has choppy audio with long silence gaps — audio analysis of a recording shows 16 silence gaps, some up to 3.7 seconds long. Browser test calls on the Retell dashboard work perfectly every time. This started a few weeks ago with no changes on our side.

Details:

  • Agent ID: agent_ed77961ce3b9e79decd236d6a6
  • Agent name: Claire SMS
  • Voice: retell-Rita
  • Example call ID: call_f86f8f2f36c7b3e4d3d2a9bb3c7
  • Trigger method: REST API (create-phone-call)
  • Phone number purchased directly through Retell

What we’ve ruled out:

  • Audio encoding — recording is 24kHz PCM, correct for Retell
  • Our server — packet capture shows 0 RTP streams on our end, audio path is entirely within Retell infrastructure
  • Agent config — dynamic variables passing correctly, calls complete and analyze successfully, latency looks normal in logs

Has anything changed on your infrastructure in the past few weeks that could affect PSTN call audio quality? Happy to share the recording if helpful.

2

Hello @steve.k

I’ve passed this issue on to our team for further investigation.

We’ll keep you updated as soon as we have more information.

Best regards

3

Hi, is there an update here?

4

Hey @steve.k

It appears this may be a TTS-related issue with the Retell voice. I’ve forwarded this to our product team for further investigation.

We’ll keep you updated as soon as we have more information.

5

Hi @steve.k

Team looked into it and there are 2 possible issues here:

  • Team noticed that Claude models occasionally have gaps in the middle of streaming responses. They might send part of a sentence, have a few seconds delay, then send the rest of the sentence. This can cause there to be pauses in the agent’s response. Switching to GPT 4.1 would help in these cases.

  • Some MiniMax voices can pause at weird places if the voice speed is too low. Increasing the voice speed would also help.

Thank You