WebRTC powers browser-based calls, and it is now a common way to connect a voice AI agent to a customer's browser or phone app. It normally runs over UDP, because for live speech a late packet is useless and it is better to skip it than to wait. A jitter buffer, loss recovery and congestion control then keep the audio usable on an unreliable network.
The hard part is everything around the media. Two devices behind firewalls need STUN and TURN servers to find a route, group calls usually need a media server, and all of it needs operating. Most companies should use a managed provider rather than run this infrastructure themselves.