SIP One-Way Audio: A Step-by-Step Troubleshooting Guide
One-way audio on SIP calls almost always means RTP is not reaching one side. Check SDP addresses, NAT, SIP ALG and firewall ports in that order to fix it.
Réponse rapide
One-way audio on a SIP call means signalling worked but RTP media flows in only one direction. Fix it in this order: read the SDP c= and m= lines for private addresses, switch off SIP ALG and double NAT, open the RTP port range both ways and check carrier IP allow-lists, use symmetric RTP or ICE where a NAT sits in the path, then capture packets on each hop.
A call that rings, connects and then leaves one person talking to silence is the most common media fault in SIP. It is also the most misdiagnosed, because the signalling looks perfect: the INVITE was answered, the call shows as connected, billing starts. Everything that went wrong happened somewhere else, in the separate stream of audio packets that SIP does not carry.
This guide gives a fixed diagnosis order, so you stop guessing. It is written for outbound teams running a cloud dialer with a carrier they bring themselves, where audio crosses three legs: the agent leg, the media node in the middle and the carrier leg. The fault is almost always on exactly one of them.
What one-way audio actually means
SIP (RFC 3261) is a signalling protocol. It sets up, changes and ends sessions, and it does not carry the audio itself. The audio travels separately as RTP, which usually runs over UDP and puts sequence numbers and timestamps on each packet (RFC 3550). The two sides agree where to send that RTP inside the SIP message body, using SDP.
So "the call connected" only proves the signalling path works. One-way audio means one of the two RTP streams, caller to callee or callee to caller, is being sent to the wrong place, blocked on the way, or never started. Which direction fails tells you where to look:
| Symptôme | What it usually means | Start at |
|---|---|---|
| Customer hears the agent, agent hears nothing | The stream from the far side never reaches you: wrong address in the far side's SDP, or your firewall or NAT drops inbound RTP | Steps 1, 2 and 3 |
| Agent hears the customer, customer hears nothing | Your SDP advertised an address the far side cannot reach, or the carrier only accepts RTP from allow-listed IPs | Steps 1, 3 and 4 |
Before anything else, make three test calls and write down which direction fails each time.
Step 1: Read the SDP
The offer/answer model in RFC 3264 is the whole story of where audio goes. In an answer, the connection address (the c= line) and the port on the m= line together say where the answerer wants to receive media. Each side sends its RTP to the address and port the other side wrote down. If one side wrote down an address that the other cannot reach, that one direction goes nowhere.
If every destination fails the same way, suspect your own network; if only one route fails, suspect that route. Take a SIP trace of one failing call and find the INVITE and the 200 OK. Read three things in each body:
- The c= address. A private address such as 10.x.x.x, 172.16-31.x.x or 192.168.x.x in a message that crossed the public internet is the classic cause. The far side will try to send audio to an address that does not exist for it.
- The m= port. A port of zero rejects that stream, so neither side sends media for it. A port outside the range your firewall opens is just as fatal.
- The direction attribute. Streams are
sendrecvby default.sendonly,recvonlyandinactiveare legal in the standard, and an answer must be compatible with the offer. A stream stuck atsendonlyorrecvonlyafter a hold and resume is a real source of one-way audio that no network change will fix.
The RTCP stream, which carries reception statistics, should go to the next higher (odd) port by default (RFC 3550, section 11), so open the port pair, not the single port. Also compare the address in the SDP with the source address of the RTP packets you actually receive. When they differ, a NAT or a media proxy is in the path, and Steps 2 and 4 matter.
Step 2: Find SIP ALG and double NAT, then switch them off
Many office routers and some firewalls ship with a feature called SIP ALG (application layer gateway). It tries to be helpful by rewriting the addresses inside SIP messages so they match the router's public address. In practice it often rewrites them wrongly, rewrites some messages and not others, or breaks a re-INVITE. The result is exactly the picture in Step 1: the SDP the far side receives does not match the packets it sees.
Check for it, in this order:
- Compare the SDP you sent with the SDP as received on the far side. If the c= address changed in transit and you did not change it, something in the path rewrote it.
- Look for a "SIP ALG", "SIP helper" or "SIP passthrough" setting on the router or firewall, and disable it for a test call. Then re-test.
- Check for double NAT: a router behind another router, for example an ISP box feeding your own firewall. Traffic that is translated twice needs the outer device to forward RTP to the inner one. The simplest fix is to put the inner device in bridge mode or move the media to a host with a public address.
In a cloud dialer topology this step mostly concerns the agent's own office or home network and any on-premise PBX you connect. The media node and the carrier typically have public addresses, which is why the fault so often lives on the customer side of a NAT.
Step 3: Open the RTP port range and check allow-lists
Signalling on port 5060 or 5061 passing does not mean a single RTP packet will. RTP uses a separate, much wider UDP port range, and firewalls that were opened for SIP only will block it. Check both directions:
- Outbound from you: your firewall must allow UDP out to the far side's media addresses on its negotiated port.
- Inbound to you: the media node needs the whole configured UDP range open to the carrier's media IPs. On a firewall that keeps state, return traffic is only admitted if it matches a flow you started, which is why inbound-only rules and NAT timeouts produce one-way audio that works for a few seconds and stops.
- The carrier's allow-list. Carriers that authenticate by IP often allow-list signalling and media addresses separately. If your media comes from an address you did not give them, the carrier may accept the call and silently drop your RTP. Re-send the carrier the full list of signalling and media addresses.
The gateway form in DialerBee shows an RTP port range on its Media tab (the form default is 16384 to 32768). Whatever range your platform is configured with is the range the carrier and every firewall in between must allow.
Step 4: Symmetric RTP, rport, STUN and ICE
When a NAT sits between you and the far side, there are four tools, and each solves a different problem. Pick by what is broken, not by habit.
| Tool | Ce que cela fait | When it applies |
|---|---|---|
| rport (RFC 3581) | Asks the server to send SIP responses to the actual source port of the request, so replies match the NAT mapping | Signalling replies not arriving from behind a NAT. It fixes SIP, not RTP |
| Symmetric RTP (also called comedia or latching) | The media side sends RTP back to wherever the first packets came from, ignoring the SDP address | The far side is behind a NAT and advertised a private address in its SDP |
| STUN (RFC 5389, since replaced by RFC 8489) | A client learns the public address and port its NAT assigned. The RFC stresses that STUN alone is not a complete NAT solution | An endpoint must put its public address into its own SDP. Works with simple NATs only |
| ICE (RFC 8445) | Both peers gather candidate addresses (host, server-reflexive, relayed), test pairs with connectivity checks and pick a working one | Browser (WebRTC) endpoints and any peer pair where direct paths are uncertain |
Two practical rules follow. First, rport does not repair RTP: a call with perfect signalling and no inbound audio is not an rport problem. Second, STUN on its own fails behind NATs that map a different external port per destination, which is when you move to ICE with a TURN relay as the fallback.
Step 5: Capture packets on each hop and compare directions
When reading the SDP does not settle it, stop theorising and capture. One capture on the media node, plus one on the agent side or at the carrier edge if you can get it, tells you which hop loses the packets.
- Start a capture on the media node's public interface filtered to UDP, for example
tcpdump -ni any udp portrange 16384-32768 -w oneway.pcap, using the range your platform is configured with. - Place a single test call and speak on both ends, so both directions carry sound rather than silence suppression.
- Open the capture in a packet analyser and list the RTP streams. Look for two streams per call, one in each direction.
Then compare what you see:
- Only one stream present. The missing direction never reached this hop, so the loss is upstream of the capture point. Move the capture one hop outward and repeat.
- Both streams present on the media node, one side hears nothing. The packets are leaving but not arriving. Compare the destination address of the outgoing stream to the address in the far side's SDP. A mismatch is Step 1 again.
- A stream with the right address but ICMP "port unreachable" coming back. The destination host is not listening on that port: the far end's media process or the media node's range is misconfigured.
RTCP reception reports, defined in RFC 3550, help here. A receiver includes a report block only for sources it has received RTP from since its last report, so if the far side's RTCP carries no report block for your stream, or a high fraction lost, your packets were sent and are not arriving.
Silent recordings versus one-way audio
A related fault looks similar and has a different cause. If calls sound fine to both parties but the recording is silent or has only one voice, the live call is healthy and the recording leg is not. This happens when media bypass (direct media) is on: RTP flows directly between the two ends and never touches the node that records. Check whether media bypass is enabled on the gateway. Keeping it off keeps media on the platform, which recording depends on, and the enregistrement d'appel feature relies on that path.
Browser (WebRTC) agents: what changes on the agent leg
When an agent works from a browser softphone, the agent leg is WebRTC. That changes the toolbox:
- ICE is mandatory. The browser gathers candidates and runs connectivity checks, so STUN and, where direct paths fail, a TURN relay do the NAT work for you.
- Permissions matter. A denied or wrong microphone looks exactly like one-way audio. If the customer hears nothing, check the browser microphone selection before touching the network.
DialerBee's agent desktop includes a media pre-flight that checks the microphone, the speakers and the media path before an agent goes ready, so a broken headset or a blocked path is found before a customer is on the line. For the wider trade-off between browser and desk-phone agents, read WebRTC vs SIP phones et le WebRTC softphone glossary entry.
Comment DialerBee s'intègre
On a bring-your-own-carrier setup you keep the trunk and the carrier relationship, and the media settings live on the gateway. The fields that matter for this problem are on the Media tab of the gateway form: NAT Traversal (None by default, with STUN, TURN and ICE), the RTP Port Range, SRTP Mode and Media Bypass / Direct Media (off by default). On the General and Authentication tabs you set the transport and whether the carrier trusts your IPs or your credentials. The full walk-through is in Comment connecter votre trunk SIP (configuration BYOC) et le page de présentation BYOC describes the model.
Which deployment pattern you use decides who owns which leg. If DialerBee supplies the voice, the carrier leg is managed for you. If you keep your own PBX or bring your own carrier, the media path between your infrastructure and the platform is yours to keep open. The deployment patterns page explains the three options. Whichever you pick, test two-way audio on every new trunk with real calls before a campaign goes live, as the setup guide's test plan recommends.
One-way audio checklist
- Reproduce with three test calls to different destinations and note which direction fails.
- Pull the INVITE and 200 OK. Check c= addresses for private or zero values.
- Check m= ports and direction attributes on every stream.
- Disable SIP ALG on every router and firewall in the path. Remove double NAT.
- Open the full RTP UDP range in both directions, both rtp and rtcp ports.
- Confirm the carrier has your signalling and media IPs on its allow-list.
- Enable symmetric RTP or ICE where a NAT sits on the far side.
- Capture on the media node and compare the two RTP directions.
- If recordings are silent but calls are fine, check Media Bypass.
- For browser agents, check microphone permission, then UDP reachability for ICE.
Foire aux questions
What causes one-way audio on SIP calls?
RTP media reaching only one side. The usual causes are a private or wrong address in the SDP, SIP ALG rewriting messages, a firewall that blocks the RTP port range in one direction, or a carrier that accepts media only from allow-listed IPs.
Why does the call connect if there is no audio?
SIP signalling and RTP media are separate flows. The call is considered connected when the SIP exchange completes, while audio depends on a second set of UDP packets that can be blocked or misaddressed independently.
How do I fix one-way audio on a SIP trunk?
Read the SDP c= and m= lines in the INVITE and 200 OK, disable SIP ALG, open the RTP port range in both directions, give the carrier every media IP you send from, enable symmetric RTP or ICE where a NAT is involved, then capture packets on each hop to find where one direction disappears.
What is SIP ALG and should I disable it?
SIP ALG is a router or firewall feature that rewrites addresses inside SIP messages. It often rewrites them incorrectly, so a common first step on any one-way audio fault is to disable it for a test call and see whether the problem disappears.
What is the difference between rport, STUN and ICE?
rport (RFC 3581) makes SIP responses return to the request's real source port. STUN (RFC 5389) lets a client discover its public address and port. ICE (RFC 8445) gathers several candidate addresses, tests pairs and picks one that works. Only STUN and ICE help media, and STUN alone is not a complete solution.
Why are my call recordings silent when the call sounded fine?
Most often media bypass or direct media is on, so RTP never passes through the node that records. Turn it off for recorded campaigns, or confirm the recording path with a test call before relying on it.
Sources
- RFC 3264, An Offer/Answer Model with the Session Description Protocol (consulté le 7 novembre 2026)
- RFC 3550, RTP: A Transport Protocol for Real-Time Applications (consulté le 7 novembre 2026)
- RFC 3581, An Extension to SIP for Symmetric Response Routing (consulté le 7 novembre 2026)
- RFC 5389, Session Traversal Utilities for NAT (STUN), now obsoleted by RFC 8489 (consulté le 7 novembre 2026)
- RFC 8445, Interactive Connectivity Establishment (ICE) (consulté le 7 novembre 2026)
- RFC 3261, SIP: Session Initiation Protocol (consulté le 7 novembre 2026)
Lectures complémentaires
- Comment connecter votre trunk SIP (configuration BYOC)
- Qu'est-ce qu'une ligne SIP ?
- Téléphones WebRTC vs SIP
- Apportez votre propre opérateur
- Enregistrement d'appel
- Modèles de déploiement
- SIP 488 Not Acceptable Here: fixing a codec mismatch
- DTMF not working on a SIP trunk? How to fix keypresses
- SIP error codes on outbound calls: 403, 486, 503 and more
- Session Border Controller (SBC)
- Jitter (VoIP)
À lire aussi : WebRTC Agent Audio Problems: A Fix Guide · What Is RTP (Real-time Transport Protocol)?
À lire aussi : What Is NAT Traversal? STUN, TURN and ICE
Articles connexes
Prêt à voir DialerBee en action ?
Demandez une démo en direct de 15 minutes, ou démarrez un essai gratuit de 14 jours — sans diaporama, sans engagement. Les appels démarrent dès que votre propre trunk SIP ou PBX est connecté et approuvé.
Essai gratuit de 14 jours · crédit IA gratuit · sans carte · 11 langues · BYOC · contrôles qui contribuent à la conformité