Hi Nicolas, thanks a lot for your message — really appreciate your input!
On my side, I'm still in the testing phase, mainly because of a major audio quality issue between Vapi and Telnyx (using Programmable Voice). Like you, I often see calls ending as “Silence Timed Out” or similar. In my case, it’s simply because the AI **can’t understand what I’m saying**: the incoming audio on Vapi’s side is so degraded (robotic, choppy, unintelligible) that it can’t transcribe anything, and therefore doesn’t respond. Eventually, the system hangs up.
Occasionally, after several attempts, a keyword does manage to get through, which allows the AI to trigger a call transfer — that’s actually what I’m trying to test. But it only works one out of two or three times, and only if I repeat the keyword five or ten times. So it’s clearly an audio input issue, not a logic problem on the AI’s side.
I also tested the SIP integration (I believe you're referring to SIP Trunking), and just like you, audio quality is perfect there — no issues at all. The problem for me is that call transfers don’t work in that setup. I need to transfer calls to a landline number in France (e.g., a restaurant), and with SIP Trunking, the call instantly disconnects when the transfer is triggered — as if nothing happened.
So even though SIP Trunking gives much better audio quality, the lack of working call transfers forced me to put it aside for now. That’s why I’m sticking with Programmable Voice, which handles transfers properly… but has this major issue with incoming audio to the assistant.
Thanks again for your message — if you’ve got any other tips or experiences, I’d love to hear them!