Yeah, that explains exactly what you’re seeing classic codec mismatch.
Since VAPI is forcing PCMU, the only reliable fix right now is to handle transcoding before the audio reaches them. I’ve solved this by placing a lightweight SIP layer (like FreeSWITCH/Asterisk) in between to convert PCMA → PCMU, which stabilizes the audio and fixes ASR immediately.
Happy to help you set that up or test a cleaner routing path if you want.
@eytanlevy