some users and —I remember myself experiencing this once a while as well— reported that the AI assistant's voice breaks off like there would be lag or internet issues. what the reasons could be to this happen? trying to understand how to mitigate these issues.

To address the assistant's voice breaking off, try adjusting the

stopSpeakingPlan

to use fast Voice Activity Detection (VAD) for interruption handling. This can make the assistant more responsive and reduce unwanted silences:

json
"stopSpeakingPlan": {
  "numWords": 0,
  "voiceSeconds": 0.2,
  "backoffSeconds": 1.0
}

Apply this in your assistant's configuration and test in your environment to ensure it balances responsiveness and noise sensitivity. For more details, see the [Voice pipeline configuration documentation](https://docs.vapi.ai/customization/voice-pipeline-configuration). Source: - [Voice pipeline configuration](https://docs.vapi.ai/customization/voice-pipeline-configuration)

Are you getting this error in the UI testing?

@Operations📟 you mean in the vapi dashboard? no, I'm using the Vapi web sdk

Oh ok interesting. I disscovered a majot bug that is causing really random and somtimes bizarre and unexpected glitches. VAPI no logner interprets slanted " and ' https://cdn.discordapp.com/attachments/1463292434439078063/1463294543985246228/Photos_9Zp9EcGlu7.png?ex=69714ebd&is=696ffd3d&hm=7301bcb24836e9b72e2660e9a03eed478f72851715f359ca67eef8afe6965cdd&

for the prompts? I think this is not related to my original issue

My AI voices broke and the only common denominator was that the secret update that changed the prompt parsing for slanted " and '. My AI's were meowing growing and releasing hellish noises. Not I fixed my prompt ' and " symbols and I don't have any problems.

Hi adrivelasco, Thanks — two things to try based on what you described: 1. Prompt-parsing bug • Your discovery about slanted quotes causing corrupt prompts matches a known recent break: replace any “curly”/slanted quotes/apostrophes with straight ASCII " and ' in your prompt text before sending. That fixed the meowing/strange hallucinations for other customers. If you edit prompts in your web SDK payloads or files, normalize quotes (e.g., via a small replace step). 1. Voice cutting / mid-sentence silences • Likely causes: VAD sensitivity, TTS/provider issues, or TTS latency. Quick mitigations: • Make stopSpeakingPlan less sensitive (increase voiceSeconds) to avoid false VAD clears: •

"stopSpeakingPlan": { "numWords": 0, "voiceSeconds": 0.4, "backoffSeconds": 1.0 }

• If you prefer faster interruption handling, decrease voiceSeconds (e.g., 0.2) but watch for premature cuts. • If using a third-party TTS (ElevenLabs etc.), test switching provider or use a fallback voice in assistant config. • Ensure model/timeouts aren’t truncating output (increase Max Tokens / timeout if you set them low). If you want, share a representative Call ID where the bot meowed or cut out and the timestamp and I’ll check the call logs for whether it was prompt parsing, VAD, or a TTS/provider error.