When I try GPT realtime for my assistant, it works fine on Web, but when I call the Vapi number on my phone, the voice is sound like it's on a slow motion and it's robotic. But if you listen to the call log after, it sounds fine! sample call ID: 9e0ccca2-5848-4c16-962b-f12299b6ca54

Suggestion: Robotic or slow-motion audio on phone calls is often caused by a mismatch in audio format or sample rate. Make sure your assistant's configuration matches the required PCM audio specs for telephony:

json
"voice": {
  "provider": "openai",
  "voiceId": "alloy", // or another supported voice
  "sampleRate": 24000 // Must match the telephony requirement
}

Double-check that your assistant is set to output raw PCM, mono, 16-bit little-endian audio at the exact sample rate requested by Vapi. Even a small mismatch can cause distorted playback on phone calls ([see documentation](https://docs.vapi.ai/customization/custom-voices/custom-tts#audio-generation)). Source: - [Custom TTS integration](https://docs.vapi.ai/customization/custom-voices/custom-tts)

@User can you respond please?

i have r=the same issue, and dont see the sample rate in the code or UI in my plan. I had to change models to fix this

customer support seems to be 404!

i switched models, I dont think theres a fix

Hey! To help track down this issue, could you share: \- When exactly this happened (the timestamp) \- What response you expected to get \- What response you actually got instead This would really help us figure out what went wrong!

Hi Praveen, This issue started about 10 days ago for me, when I wanted to test and switch to GPT-Realtime model. but I can still reproduce the same issue right now. - What Happens: when I use GPT-Realtime or GPT 4o Realtime model for my assistant, the voice on the phone is distorted and robotic. - Expected Response: I'd expected to hear the same voice and tone on the phone, as when I call on the web - While the web based call sounds as expected, AI's voice on the phone call sounds robotic and distinctly off in comparison

I just had the same problem as described above and confirmed that sampleRate is not an option in the API.

also having this issue

same here

@User please respond

same issue here. using openai non-real time model, everything is good. using openai real-time model, web = okay using twilio integration = okay using free vapi number = distorted using vonage integration = distorted

Doesn't work with a free VAPI number but works when using a Twilio number

I’m also having this issue. Would love to know the status on this

I have another work around when using Vonage. Instead of using Vonage integration provided by VAPI, On your Vonage website, Use Voice -> Forward to Phone. Webhook: https://api.vapi.ai/twilio/inbound_call Then, it works. My guess is natively, there is conflict between sampling rate, thus sound slowed down. But, there is sampling conversion using webhook? just a guess.

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.

No progress. still the same issue.

Hey, Can you send me the latest call_id? Along with that, can you provide me the duration in the recording where you had this issue?

Just FYI -- this is still a problem. Call ID = 019c5377-5d06-7aa5-8668-523556fdd5d8 The recording in the Call Logs is fine, but the voice on the actual call is like talking to a giant that had too much to drink the night before. Also, there's a slight lag before the model starts speaking. Lastly, it doesn't seem the Realtime is any faster than any of the other models, even though it shows "550ms" as the estimated latency. This is going from cell phone to Vapi-hosted #.