Skip to content

OBS

Jump to bottom
Shannon Atkinson edited this page Aug 27, 2026 · 1 revision

Generated from docs/OBS.md. Edit that file and re-run node scripts/publish-wiki.mjs --push. An edit made here is a fork of the documentation that nothing reconciles, and the next run of this script will overwrite it without asking.

OBS setup

polyemesis accepts a stream from anything that can push MPEG-TS over SRT or a single track over RTMP. OBS is what most people use, so it gets its own page.

On Enhanced RTMP. OBS 30.2+ can send several audio tracks over RTMP using the Enhanced RTMP multitrack format. FFmpeg gained multitrack FLV demuxing in late 2024, so a new enough build carries those tracks through polyemesis's existing ingest command unchanged — verified up to six tracks on FFmpeg 8.1. Ubuntu 24.04's stock FFmpeg 6.1.1 cannot: it refuses with "at most one audio stream is supported in flv".

OBS cannot either, and that is now measured rather than open. OBS 30.2.3 was run headless against a real polyemesis with three tracks configured, each on its own mixer, StreamMultiTrackAudioMixes=7 and a custom RTMP service — and the captured wire bytes are 0xaf legacy ×3541 with no multitrack tag anywhere. The path is gated on supports_additional_audio_track, which no service in OBS's services.json declares (0 of 91), so it is unreachable even for custom RTMP. A weekly job re-checks it. If your encoder is OBS, use SRT. See evidence/enhanced-rtmp-multitrack.md.

Only one of these configurations unlocks per-destination audio routing: multitrack over SRT. The others work, and are documented here so that choosing one is deliberate rather than accidental.

Multitrack over SRT

OBS sends several audio tracks in one MPEG-TS stream over SRT, and polyemesis gives each destination its own mix of them.

0. Create the source you are pushing to

A fresh install has no source, and YOUR_TOKEN below is a source's publish token — so there is nothing to point OBS at until one exists. Add one on the Sources page; a name is all it asks for. That page then shows the whole publish URL, token included, which is the value to paste into the File path or URL field in step 3.

1. Enable the audio tracks

Settings → Output → Audio (Advanced mode). Set a bitrate for each track you intend to use — 160 kbps for tracks 1–3 is a reasonable start.

2. Assign sources to tracks

In the Audio Mixer, click the gear icon → Advanced Audio Properties. Each source has six Tracks checkboxes. A typical layout:

Source 1 2 3 Meaning
Desktop audio (music) ✓ full mix only
Game audio ✓ ✓ full + clean
Microphone ✓ ✓ ✓ everywhere

Track 1 = everything, track 2 = clean (no music), track 3 = mic only.

3. Configure the Custom FFmpeg output

Settings → Output → Output Mode: Advanced → Recording tab, set Type: Custom Output (FFmpeg), then:

Field Value
FFmpeg Output Type Output to URL
File path or URL srt://YOUR_SERVER:6000?mode=caller&transtype=live&latency=200000&streamid=YOUR_TOKEN
Container Format mpegts
Muxer Settings (leave blank)
Video Bitrate e.g. 6000 Kbps
Keyframe interval (frames) 60 (2 s at 30 fps)
Video Encoder libx264 (or h264_nvenc, h264_videotoolbox)
Video Encoder Settings preset=veryfast tune=zerolatency
Audio Bitrate 160 Kbps
Audio Encoder aac
Audio Track tick 1, 2, 3 (up to 6)

latency is in microseconds. 200000 is 200 ms, and it must match the latency set on that source's own ingest, on the Sources page. (Settings → Ingest edits the same numbers for the default source, and on an install with no source at all it has nothing to edit and says so.) Writing 200 gives you a 0.2 ms buffer and a stream that falls apart on the first jitter.

streamid is the address, not an extra. One SRT port serves every source and they are told apart by their publish token, so a publisher that presents no token — or an unrecognised one — is refused rather than guessed at. Copy the whole URL from Sources, which fills the token in for you. Rotating a token keeps the old one working for five minutes, so you can change it without cutting a live stream. See DESIGN-ONE-PORT-ONLY.md for why.

Copy the exact URL from the polyemesis dashboard rather than assembling it by hand — it renders your server's hostname and current settings, including the passphrase if you set one.

Passphrases use letters, digits and - _ . ~. Anything else is refused when you set it, and the reason is worth knowing: the passphrase travels in the URL above, and an encoder does not un-escape it. A ; would be written %3B in the URL, sent as the literal text %3B, and compared against the ; you stored — so a correct passphrase would be refused on every connection, with nothing on screen to say why.

4. Start it

Press Start Recording, not Start Streaming. With Output to URL, OBS's "recording" is the SRT push.

5. Verify

The polyemesis dashboard should show your track count and video format. Open Audio meters, play music, and confirm track 1 moves while track 2 stays flat.

That last check is the whole product working. Do it before you go live rather than after — a "clean" track that is not actually clean is the failure mode this software exists to prevent, and it is silent until somebody files a copyright claim.

Standard RTMP (single track)

For encoders that cannot do SRT. Classic RTMP carries one audio track, so per-destination track routing does not apply — though gain, matrix panning and 5.1 downmix still do.

In polyemesis: open the source under Sources and set Ingest → Mode: RTMP. It gets its own app name and stream key. In OBS: Settings → Stream, Service Custom...:

Field Value
Server rtmp://YOUR_SERVER:1935/live
Stream Key that source's stream key, from Sources

Then press Start Streaming.

The stream key is the address, not an extra. One RTMP port serves every source and they are told apart by the key in the publish URL, exactly as SRT tells them apart by token — so a publisher presenting an unrecognised key is refused rather than landing on whichever source happens to be there. Copy the Server and Stream Key from Sources rather than assembling them by hand: each source has its own pair, and pasting one source's key under another source's name is the mistake this shape makes easy to notice and easy to fix.

Any number of sources may use RTMP ingest, the same as SRT. A second RTMP encoder is an ordinary second source; it needs no extra port and no configuration beyond its own stream key. See DESIGN-ONE-PORT-ONLY.md for how that came to be true — until 2026-08-06 an install could carry exactly one RTMP source.

What RTMP still does not give you is per-destination track routing, which is a limit of the format rather than of the listener. If your encoder can speak SRT, use SRT.

Enhanced RTMP multitrack (works on FFmpeg 7.1+, not from OBS)

OBS 30.2+ can send multiple audio tracks over Enhanced RTMP/FLV.

It works on a new enough FFmpeg. Multitrack FLV demuxing landed in FFmpeg 7.1, and from there the tracks arrive through polyemesis's existing ingest command unchanged — verified end to end: a destination configured for tracks 1 and 3 received exactly those two and neither of the others. It does not work on FFmpeg 6.1.1, which is Ubuntu 24.04's stock build: that refuses with "at most one audio stream is supported in flv", and the extra tracks are lost with no error at either end.

What has not been done is a run with OBS itself as the publisher — the testing used FFmpeg. OBS writes the same MULTITRACKTYPE_ONE_TRACK format (read from its flv-mux.c), so the wire format is not in question, but the handshake and metadata path are unconfirmed. SRT remains the operated path.

See evidence/enhanced-rtmp-multitrack.md.

config.yaml used to declare an enhancedRtmp key. It has been removed: it was an inert placeholder that read as a switch, and it was kept on the belief that old config files needed it to keep parsing. They do not — unknown keys are ignored — so the key is gone and a config that still names it loads fine.

For multiple audio tracks, use SRT ingest.

See also

polyemesis

Getting it running

The routing

Operating it

Automating it

Understanding it

Clone this wiki locally