I bought a number from the Dominican Republic on Telnyx, and when I try to do an outbound call, the bot doesn't pick up correctly the voice of the person. Like there's a hugeeeee lag between what the AI is picking and the person talking. I reached out to Telnyx and they say there's an issue on the VAPI side. I'd like to have more insight because I need to promote it this asap.

Shorten the assistant prompt, disable the analysisPlan, use eleven_flash_v2_5 instead of multilingual, and add the endpointing settings down below

{
  "startSpeakingPlan": {
    "transcriptionEndpointingPlan": {
      "onPunctuationSeconds": 0.1,
      "onNoPunctuationSeconds": 1.2,
      "onNumberSeconds": 0.4
    },
    "waitSeconds": 0.3
  },
  "stopSpeakingPlan": {
    "numWords": 0,
    "voiceSeconds": 0.2,
    "backoffSeconds": 0.8
  }
}

@Ramphis reyes I've listened to the recording you sent - the static you are hearing is generated due to several factors: 1. CNG Generation the remote destination - It seems like the Comfort Noise Generator being used at the remote end doesn't kick in properly, potentially at the Dominican Republic. 2. Audio distortion - For VAPI (or any other voice AI engine) to work properly, it needs to rely on a link capable of providing a solid 16Khz audio stream, over a 64kbps link - what I've heard from the recording is audio encoding using a g723.1 codec - basically, a 6.3kbps codec using 8Khz audio stream - the AI can't use that. Traditional telephony uses 8Khz and 64kbps bandwidth. What you need is something that will re-sample the inbound audio and up-scale it on the fly. If you want, I'll be happy to see if using Cloudonix as an SBC and audio upscaler will work for you - as this is something the platform does automatically. Feel free to DM me for more details.

Let us know how your testing goes and how you would like to proceed.