A browser video call and a phone-system call can describe the same conversation while speaking different protocol languages. WebRTC uses browser-oriented session procedures, ICE connectivity checks, and a DTLS-associated media path; SIP establishes sessions and uses SDP to describe media formats and addresses. Connecting them requires a gateway or equivalent interworking function unless the endpoint already implements both sides of the required profile.
Make that boundary explicit in the architecture. Let the browser-facing side handle WebRTC session establishment and media security, let the phone-facing side handle SIP and its negotiated media profile, and keep the translation point observable and deliberate. That makes codec failures, recording access, and trust boundaries design decisions rather than side effects. Voice agents on a phone line raise a separate channel question, covered in Voice Agents on a Phone Line vs in the Browser; the mechanics below apply to human calls and video.
SIP WebRTC interop starts with two separate jobs
The first job is signalling. SIP establishes, modifies, and terminates a session. SDP travels with that signalling and describes the media streams, including formats and network addresses. SIP and the media session can take different network paths, so a SIP request does not, by itself, tell you where the audio or video packets will travel. RFC 3261 defines this separation.
The second job is media interworking. The browser and the phone system must agree on a usable RTP session, security context, payload format, and stream structure. A gateway may translate the signalling state, rewrite SDP, create separate RTP sessions on each side, and adapt the media when the two profiles do not line up. RFC 8872 describes this kind of RTP gateway interworking.
That division gives the implementation a useful boundary: SIP messages and SDP describe the call, while the gateway makes the negotiated media path work. The same boundary sits under browser-to-contact-centre designs such as IVR-less voice and video calling for customer support. The next issue is how the browser reaches that boundary securely.
What the browser side brings to SIP WebRTC interop
WebRTC connectivity uses ICE candidate pairs. ICE gathers possible paths and checks candidate pairs for connectivity; DTLS is associated with the ICE-established component that carries the secured media relationship. RFC 8835 defines this relationship for WebRTC transports.
The browser also follows the session-description procedures defined by JSEP. JSEP requires the exchange of session descriptions through the SDP offer/answer model, but it does not define the signalling protocol that transports those descriptions. RFC 8829 therefore leaves the application to carry the offer and answer between the browser and the gateway. That transport could be a WebSocket-based application channel, an HTTP exchange, or another signalling path chosen by the product. WebRTC Architecture: Everything That Runs Between Two Video Tabs maps the rest of the browser-side stack.
The gateway has to receive the browser's offer, decide how it maps to the SIP side, and return a browser-compatible answer. It may also need to manage ICE candidates and the timing of session changes. SIP does not automatically become the browser's signalling channel simply because the far end is a SIP endpoint.
This browser boundary is the first place to observe failures. A call can fail before media starts because the application did not exchange descriptions, because the SDP mapping was invalid, or because no candidate pair could connect. The separate NAT traversal guide for WebRTC covers the connectivity problem in greater depth.
How SIP and WebRTC negotiate codecs
Codec negotiation happens through SDP offer and answer. The offer proposes media formats, and the answer identifies the formats the other side accepts. If the answerer has no common format for a stream, that stream must be rejected under RFC 3264.
WebRTC video has a defined baseline of implementation requirements: browsers and non-browser endpoints that support video must implement VP8 and H.264 Constrained Baseline. RFC 7742 establishes those requirements. They describe what an implementation must support; they do not determine which codec a particular call selects.
Selection still depends on the SDP offer and answer, codec profiles, packetization parameters, endpoint capabilities, and policy. A browser may offer H.264 with parameters that the phone-side endpoint cannot accept, or the SIP side may advertise a format that the browser does not include. The gateway should expose the rejected media stream and the exact SDP cause rather than reducing every failure to “codec mismatch.”
When both sides share compatible codecs, payload formats, multiplexing choices, and RTP/RTCP capabilities, the interworking layer may translate session details without decoding and re-encoding the content. When they do not, transcoding becomes an interworking function. RFC 8872 identifies transcoding, video rescaling, and changes to redundancy or scalability layers as possible gateway functions.
The implementation decision is therefore straightforward: negotiate a common format first, and transcode only when the deployed endpoints require it. Transcoding adds processing and creates another media operation to monitor, while a failed negotiation ends the affected stream altogether. The media-security profile determines whether either path can be used without terminating protected media.
Where the media security boundary sits
SIP deployments support different media-security profiles. Ordinary RTP is one possible media mode, while SDES-SRTP carries cryptographic parameters in SDP and relies on the enclosing signalling protocol to protect that information. RFC 4568 defines the SDP security descriptions used for SDES-SRTP. The SIP protocol model itself does not impose one universal media-security profile.
On the browser side, the gateway must work with the WebRTC security and connectivity procedures offered by the browser. On the phone side, it must create or accept the SIP endpoint's negotiated media context. If the gateway terminates an RTP session and changes protected media, it must decrypt and verify packets arriving from one context, then protect them again in the other context. RFC 8872 describes the gateway as a member of both security contexts in that situation.
That makes the gateway a trusted media endpoint for the portion it terminates. A transport relay that forwards encrypted packets without changing their protected payload has a different trust boundary. A media translator, transcoder, or recorder needs stronger access because it must inspect or change the media. RFC 5117 explains the cryptographic access required by media translators that change RTP payloads.
Document this boundary per stream and per direction. Record which component holds the relevant keying material, which component can see decoded audio or video, and which component can modify RTP. Those answers define the practical security architecture more accurately than the word “gateway” alone.
Recording changes the gateway decision
Recording changes what the gateway is trusted with. An intermediary that records protected media must operate as a man-in-the-middle or as an endpoint for the relevant security context. RFC 5763 describes that security consequence.
If the gateway already terminates media for transcoding, it may be the natural observation point for a recording function. That arrangement still requires explicit controls around decoded media, encryption at rest, retention, access, and deletion. The gateway should not be granted recording access merely because it can pass packets.
A design that preserves encrypted media through a transport relay has a different recording boundary. Recording may then require a separate media endpoint or a dedicated stream that the recorder is authorised to receive. The right choice depends on where the organisation wants the recording trust boundary and which side is responsible for the resulting media.
Recording also exposes a practical test: make a call with video, audio, hold, resume, and a session change, then inspect whether the recording contains the expected streams and timing. A successful SIP answer does not prove that the recording path has access to the media it needs.
Gateway placement determines what you can measure
Place the interworking function where it can reach both signalling domains and both media paths without hiding failures behind an opaque relay. The browser-facing side needs visibility into ICE and WebRTC session state. The phone-facing side needs SIP transaction state, SDP offers and answers, RTP and RTCP statistics, and any provider-specific response codes.
Placement also changes the path that media takes. Routing, buffering, and codec conversion occur at particular stages in that path, so latency must be measured in the deployed route rather than assigned a universal penalty. Capture timestamps at browser ingress, gateway receipt, gateway transmission, and phone-side receipt where the environment permits it.
Measure at least four outcomes:
- Session establishment: whether the browser offer, gateway translation, SIP transaction, and browser answer complete in the expected order.
- Media establishment: whether the negotiated ICE and RTP paths carry packets in both directions.
- Media continuity: whether packet loss, jitter, retransmission, key changes, hold, resume, and renegotiation produce the expected behaviour.
- Media access: whether recording, transcoding, monitoring, and policy controls can see only the streams they are authorised to process.
Keep signalling logs and media metrics correlated with a call identifier. A SIP dialog that reaches an established state while the browser receives no video is a media-path failure. The same discipline helps distinguish a rejected SDP stream from a connected call whose packets are being discarded later.
The practical interop sequence
Build the integration in this order:
- Define the endpoint contract. List the browser-facing WebRTC procedures, the SIP transport and signalling expectations, the media-security profiles, and the audio and video formats that each side can accept.
- Translate signalling. Map application session state to SIP transactions and map SDP fields in both directions. Preserve the reason for rejected streams and failed negotiations.
- Establish media. Carry out ICE and the browser-side security procedure, then create the phone-side RTP context. Treat each media leg as observable on its own.
- Resolve codec differences. Prefer a shared format. Add transcoding only for an identified incompatibility, and monitor the resulting processing and media quality.
- Place recording deliberately. Decide which trusted component can access decoded media and test recording during renegotiation, hold, resume, and video changes.
- Test failure paths. Reject incompatible codecs, block candidate pairs, interrupt the network, expire registration, and remove recording access. The system should expose a specific failure rather than leave the caller with a silent or one-way session.
This sequence keeps the gateway boundary visible while the integration is still small enough to debug. Once the call works, the same boundary becomes the place to enforce policy, observe quality, and control media access.
Bottom Line
SIP and WebRTC interop is a boundary-management problem. The browser brings ICE, JSEP, and its media-security procedures; the phone system brings SIP, SDP, and its negotiated RTP profile. A gateway or equivalent interworking function translates the session and media rules between them.
Use a shared codec where negotiation permits it, transcode only when the endpoints require it, and treat any component that terminates protected media as a trusted security endpoint. Put recording, observability, and policy at an explicitly chosen boundary, then measure the complete deployed path.
What's Next
Once a browser call reaches the phone system, the next engineering problem is keeping browser media reliable across mobile operating systems. The companion guide, Mobile Web Video Beyond iOS: What Android Chrome Does Differently, examines the Android Chrome behaviours that need their own test plan.
For the room-side architecture, the comparison of SFU, MCU, and mesh topologies covers how participant count changes media distribution.
FAQs
What is SIP WebRTC interop?
SIP WebRTC interop connects a browser media session to a SIP-based phone or communications system through a gateway or equivalent interworking function. The boundary translates session signalling, SDP, media formats, and security contexts so the two sides can establish compatible media legs.
Does WebRTC use SIP for signalling?
WebRTC does not require SIP as its signalling protocol. JSEP defines the browser-side offer/answer procedures, while the application chooses how to transport those descriptions; a gateway can carry the resulting session into SIP.
Does a SIP-to-WebRTC gateway always need to transcode video?
No. The gateway can avoid transcoding when the two sides share compatible codecs, profiles, packetization parameters, and RTP capabilities. It needs transcoding only when negotiation or media-format differences prevent the browser and SIP endpoint from using a common representation.
Can SIP and WebRTC share the same media path?
They can share a compatible media path when the endpoint and gateway support the required transport, RTP, and security profiles. In other deployments, the gateway creates independent RTP sessions for the browser and phone sides and translates between them.
Where should recording happen in a SIP WebRTC integration?
Recording should happen at a component authorised to access the relevant media security context. An intermediary recording protected media becomes a trusted endpoint or man-in-the-middle for that media, so its key access, decoded-media access, retention, and audit controls must be explicit.
Why does a SIP WebRTC call connect without video?
The video stream may have been rejected during SDP offer/answer because the endpoints had no accepted format in common, or the video media path may have failed after the session was established. Inspect the SDP result and the independent RTP and RTCP metrics for the video leg instead of treating the SIP dialog state as proof that every stream connected.
What should be tested before launching SIP WebRTC interop?
Test session establishment, codec rejection, ICE failure, one-way media, hold and resume, renegotiation, network interruption, recording access, and transcoding behaviour. Correlate browser, gateway, SIP, and RTP logs with one call identifier so each failure can be assigned to signalling, negotiation, connectivity, media processing, or policy.