Issue: Intermittent Audio Failure – LiveKit Using TURN Instead of Direct RTP (SIP ↔️ WebRTC) Description We are experiencing intermittent one-way / no-audio issues when connecting SIP calls (via rtpengine + Kamailio) to LiveKit rooms. In some calls, audio flows correctly. However, in many cases: • No RTP is received from the LiveKit side • Calls are terminated after ~20 seconds (likely due to upstream SIP timeout, in this case - WhatsApp) This behavior started suddenly without any infrastructure changes on our side. Setup Overview

WhatsApp SIP → Kamailio → rtpengine → LiveKit (Cloud) → Browser

• SIP side codec: PCMU (8kHz) • WebRTC side codec: Opus (48kHz) • rtpengine handles transcoding (PCMU ↔️ Opus) LiveKit is cloud-hosted (not self-hosted). Observed Behavior From `rtpengine-ctl call info`: ❌ Failing calls

Media (WebRTC side):
172.17.0.4:24998 <> 57.144.41.49:3484
0 packets, 0 bytes, 93 errors

• No RTP received from LiveKit side • Errors continuously increment • Call drops after ~20s ⚠️ Partially working calls

Media (WebRTC side):
172.17.0.4:25028 <> 57.144.49.49:3484
494 packets, 34 errors

• Some RTP received • Audio is intermittent ✅ Working calls

Media (WebRTC side):
172.17.0.4:23068 <> 57.144.49.49:3484
227 packets, 0 errors

• Audio works normally Key Observation In all cases, LiveKit is sending media from:

57.144.x.x:3484

This strongly suggests that: • LiveKit is using a TURN relay • Instead of a direct ICE host/srflx candidate Expected Behavior We expect LiveKit to: • Prefer direct UDP (host/srflx ICE candidates) when reachable • Avoid TURN unless strictly necessary • Even if that's not possible, how to reliably use TURN to relay media Actual Behavior LiveKit appears to: • Prefer or fall back to TURN frequently • Resulting in: ◦ No RTP flow in some cases ◦ Intermittent audio in others Impact • Calls drop after ~20 seconds due to no media • Unreliable audio experience • Production impact on SIP-based calling (WhatsApp integration) Questions 1. Under what conditions does LiveKit Cloud prefer TURN over direct candidates? 2. Is there a way to force or prioritize host/srflx candidates? 3. Could there be regional TURN routing issues causing intermittent RTP failures? 4. Are there any logs or metrics available from LiveKit Cloud to debug ICE candidate selection? Additional Notes • No changes were made to: • Kamailio • rtpengine • Network/firewall • SIP side RTP (WhatsApp ↔️ rtpengine) is always stable • Issue only occurs on WebRTC (LiveKit) side • Other SIP calls on same infrastructure are working normally What We Need • Guidance on ensuring direct RTP path (no TURN) • Debugging steps for ICE candidate selection in LiveKit Cloud • Any known issues with TURN reliability on LiveKit Cloud Optional Debug Info (can provide if needed) • Full SDP offer/answer • ICE candidate pairs • Packet captures (tcpdump) • rtpengine logs

Hi Ravi Asthana, ⚠️ This channel will be deleted soon ⚠️ Please use our new community at https://community.livekit.io for technical discussions and to post technical questions. The Slack community is not going away, but we are migrating technical questions to our new forum. Please see full details here.

Update: The call media has started working fine on its own which is weird. The quality of audio was jittery initially (on livekit's end), but has improved significantly since yesterday.