diff --git a/proposals/3291-muting.md b/proposals/3291-muting.md
new file mode 100644
index 00000000000..6fd114d5466
--- /dev/null
+++ b/proposals/3291-muting.md
@@ -0,0 +1,130 @@
+# MSC3291: Muting in VoIP calls
+
+During VoIP calls, it is common for a user to mute their microphone/camera.
+Ideally, the other side should be able to see that the opponent's camera is
+muted, so that it could reflect this in the UI (e.g. show the user's avatar
+instead of their camera feed). We would also want the changes in the mutes state
+to be quick.
+
+Using pure WebRTC there are two ways to do muting and both have their issues:
+
++ Disabling the corresponding track
++ Setting the corresponding track as `recvonly`/`inactive`
+
+The Alternatives section describes the issues with using these alone.
+
+## Proposal
+
+This MSC proposes extending the `sdp_stream_metadata` object (see
+[MSC3077](https://github.com/matrix-org/matrix-doc/pull/3077)) to allow
+indicating the mute state to the other side using the following fields:
+
++ `audio_muted` - a boolean indicating the current audio mute state
++ `video_muted` - a boolean indicating the current video mute state
+
+This MSC also adds a new call event `m.call.sdp_stream_metadata_changed`, which
+has the common VoIP fields as specified in
+[MSC2746](https://github.com/matrix-org/matrix-doc/pull/2746) (`version`,
+`call_id`, `party_id`) and a `sdp_stream_metadata` object which is the same
+thing as `sdp_stream_metadata` in `m.call.negotiate`, `m.call.invite` and
+`m.call.answer`. The client sends this event when the `sdp_stream_metadata` has
+changed but no negotiation is required (e.g. the user mutes their
+camera/microphone).
+
+All tracks should be assumed unmuted unless specified otherwise.
+
+Clients are recommended to not mute the audio of WebRTC tracks locally when a
+incoming stream has the `audio_muted` field set to `true`. This is because when the
+other user unmutes themselves, there may be a slight delay between their client
+sending audio and the `m.call.sdp_stream_metadata_changed` event arriving. If
+`enabled` is set to `false`, then any audio sent in between those two events
+will not be heard. The other user will still stop transmitting audio once they
+mute on their side, so no audio is sent without the user's knowledge.
+
+The same suggestion does not apply to `video_muted` - there clients _should_
+mute video locally, so that the receiving side doesn't see black video.
+
+### Example
+
+```JSON
+{
+    "type": "m.call.sdp_stream_metadata_changed",
+    "room_id": "!roomId",
+    "content": {
+        "version": "1",
+        "call_id": "1414213562373095",
+        "party_id": "1732050807568877",
+        "sdp_stream_metadata": {
+            "2311546231": {
+                "purpose": "m.usermedia",
+                "audio_muted:": true,
+                "video_muted": true
+            }
+        }
+    }
+}
+```
+
+This event indicates that both audio and video are muted. It is suggested the
+video track of stream `2311546231` should be hidden in the UI (probably replaced
+by an avatar). It also suggests the UI should show an indication that the audio
+track is muted but the client should not mute the audio on the receiving side.
+
+## Potential issues
+
+When the user mutes their camera, some browsers may keep sending meaningless data
+which will waste bandwidth.
+
+## Alternatives
+
+### Only disabling the corresponding track
+
+This is the solution that some clients (e.g. Element Android) use at the moment.
+While this is almost instantaneous, it doesn't allow the other side to know the
+opponent's mute state. This leads to the opponent showing a black screen for a
+muted video track and not doing anything for a muted audio track which is bad
+for UX.
+
+### Setting the corresponding track as `recvonly`/`inactive`
+
+While this would be beneficial for low bandwidth connections, it takes time. The
+delay might be acceptable for video but isn't for audio (with which you would
+assume an instantaneous mute state change). This is also problematic since there
+could be a confusion with holding (as defined in
+[MSC2746](https://github.com/matrix-org/matrix-doc/pull/2746)).
+
+### Using a separate event for muting
+
+While this might feel clearer initially, it doesn't have much real benefit. The
+mute state is in fact a meta information about the stream and using
+`sdp_stream_metadata` is also more flexible for cases where the user joins a
+call already muted. It is also more flexible in general and would be useful if
+we ever decided to do what is described in the next section.
+
+### A combination of disabling tracks, `sdp_stream_metadata` and SDP
+
+An option would be using the current method in combination with setting the
+corresponding track as `recvonly`/`inactive`. Along with this clients would need
+to set the mute state in `sdp_stream_metadata` to avoid conflicts with holding
+(as defined in [MSC2746](https://github.com/matrix-org/matrix-doc/pull/2746)).
+While this solution might be the most flexible solution as it would allow
+clients to choose between bandwidth and a mute state change delay for each
+track, it would be harder to implement and feels generally disjointed.
+
+## Security considerations
+
+None that I can think of.
+
+## Dependencies
+
++ [MSC3077](https://github.com/matrix-org/matrix-doc/pull/3077)
+
+## Unstable prefix
+
+|Release                             |Development                                  |
+|------------------------------------|---------------------------------------------|
+|`m.call.sdp_stream_metadata_changed`|`org.matrix.call.sdp_stream_metadata_changed`|
+|`sdp_stream_metadata`               |`org.matrix.msc3077.sdp_stream_metadata`     |
+
+We use an unstable prefix for `sdp_stream_metadata` to match
+[MSC3077](https://github.com/matrix-org/matrix-doc/pull/3077).