gst-meet: Integrate Jitsi Meet conferences with GStreamer pipelines

Note: gst-meet is in an alpha state and is under active development. The command-line options and the lib-gst-meet API are subject to change.

gst-meet provides a library and tool for integrating Jitsi Meet conferences with GStreamer pipelines. You can pipe audio and video into a conference as a participant, and pipe out other participants' audio and video streams.

Thanks to GStreamer's flexibility and wide range of plugins, this enables many new possibilities.

Dependencies

• gstreamer 1.20
• gst-plugins-good, gst-plugins-bad (same version as gstreamer) and any other plugins that you want to use in your pipelines
• glib
• libnice

For building:

• pkg-config
• A Rust toolchain (rustup is the easiest way to install one)

Installation

cargo install --force gst-meet (--force will upgrade gst-meet if you have already installed it.)

To integrate gst-meet into your own application, add a Cargo dependency on lib-gst-meet.

Docker

A Dockerfile is provided that uses AVStack-built Alpine APKs for gstreamer 1.20.

Nix

For nix users, a shell.nix is provided. Within the repository, run nix-shell --pure to get a shell with access to all needed dependencies (and nothing else).

Development

Install the dependencies described above, along with their -dev packages if your distribution uses them. cargo build should then successfully build the libraries and gst-meet binary.

Pipeline Structure

You can pass two different pipeline fragments to gst-meet.

--send-pipeline is for sending audio and video. If it contains an element named audio, this audio will be streamed to the conference. The audio codec must be 48kHz Opus. If it contains an element named video, this video will be streamed to the conference. The video codec must match the codec passed to --video-codec, which is VP9 by default.

--recv-pipeline is for receiving audio and video, if you want a single pipeline to handle all participants. If it contains an element named audio, a sink pad is requested on that element for each new participant, and decoded audio is sent to that pad. Similarly, if it contains an element named video, a sink pad is requred on that element for each new participant, and decoded & scaled video is sent to that pad.

--recv-pipeline-participant-template is for receiving audio and video, if you want a separate pipeline for each participant. This pipeline will be created once for each other participant in the conference. If it contains an element named audio, the participant's decoded audio will be sent to that element. If it contains an element named video, the participant's decoded & scaled video will be sent to that element. The strings {jid}, {jid_user}, {participant_id} and {nick} are replaced in the template with the participant's full JID, user part, MUC JID resource part (a.k.a. participant/occupant ID) and nickname respectively.

You can use --recv-pipeline and --recv-pipeline-participant-template together, for example to handle all the audio with a single audiomixer element but handle each video stream separately. If an audio or video element is found in both --recv-pipeline and --recv-pipeline-participant-template, then the one in --recv-pipeline is used.

Examples

A few examples of gst-meet usage are below. The GStreamer reference provides full details on available pipeline elements.

gst-meet --help lists full usage information.

Stream an Opus audio file to the conference. This is very efficient; the Opus data in the file is streamed directly without transcoding:

gst-meet --web-socket-url=wss://your.jitsi.domain/xmpp-websocket \
         --room-name=roomname \
         --send-pipeline="filesrc location=sample.opus ! queue ! oggdemux name=audio"

Stream a FLAC audio file to the conference, transcoding it to Opus:

gst-meet --web-socket-url=wss://your.jitsi.domain/xmpp-websocket \
         --room-name=roomname \
         --send-pipeline="filesrc location=shake-it-off.flac ! queue ! flacdec ! audioconvert ! audioresample ! opusenc name=audio"

Stream a .webm file containing VP9 video and Vorbis audio to the conference. This pipeline passes the VP9 stream through efficiently without transcoding, and transcodes the audio from Vorbis to Opus:

gst-meet --web-socket-url=wss://your.jitsi.domain/xmpp-websocket \
         --room-name=roomname \
         --send-pipeline="filesrc location=big-buck-bunny_trailer.webm ! queue ! matroskademux name=demuxer
                          demuxer.video_0 ! queue name=video
                          demuxer.audio_0 ! queue ! vorbisdec ! audioconvert ! audioresample ! opusenc name=audio"

Stream the default video & audio inputs to the conference, encoding as VP9 and Opus, display up to two remote participants' video streams composited side-by-side at 360p each, and play back all incoming audio mixed together (a very basic, but completely native, Jitsi Meet conference!):

gst-meet --web-socket-url=wss://your.jitsi.domain/xmpp-websocket \
         --room-name=roomname \
         --recv-video-scale-width=640 \
         --recv-video-scale-height=360 \
         --send-pipeline="autovideosrc ! queue ! videoscale ! video/x-raw,width=640,height=360 ! videoconvert ! vp9enc buffer-size=1000 deadline=1 name=video
                          autoaudiosrc ! queue ! audioconvert ! audioresample ! opusenc name=audio" \
         --recv-pipeline="audiomixer name=audio ! autoaudiosink
                          compositor name=video sink_1::xpos=640 ! autovideosink"

Record a .webm file for each other participant, containing VP9 video and Opus audio:

gst-meet --web-socket-url=wss://your.jitsi.domain/xmpp-websocket \
         --room-name=roomname \
         --video-codec=vp9 \
         --recv-pipeline-participant-template="webmmux name=muxer ! queue ! filesink location={participant_id}.webm
                                               opusenc name=audio ! muxer.audio_0
                                               vp9enc name=video ! muxer.video_0"

Play a YouTube video in the conference. By requesting Opus audio and VP9 video from YouTube, and setting the Jitsi Meet video codec to VP9, no transcoding is necessary. Note that not every YouTube video has VP9 and Opus available, so the pipeline may need adjusting for other videos.

YOUTUBE_URL="https://www.youtube.com/watch?v=vjV_2Ri2rfE"
gst-meet --web-socket-url=wss://your.jitsi.domain/xmpp-websocket \
         --room-name=roomname \
         --video-codec=vp9 \
         --send-pipeline="curlhttpsrc location=\"$(youtube-dl -g $YOUTUBE_URL -f 'bestaudio[acodec=opus]')\" ! queue ! matroskademux name=audiodemux
                          curlhttpsrc location=\"$(youtube-dl -g $YOUTUBE_URL -f 'bestvideo[vcodec=vp9]')\" ! queue ! matroskademux name=videodemux
                          audiodemux.audio_0 ! queue ! clocksync name=audio
                          videodemux.video_0 ! queue ! clocksync name=video"

Feature flags

By default, the rustls TLS library is used with the system's native root certificates. This can be turned off by passing --no-default-features to Cargo, and one of the following features can be enabled:

tls-rustls-native-roots  use rustls for TLS with the system's native root certificates (the default)
tls-rustls-webpki-roots  use rustls for TLS and bundle webpki's root certificates
tls-native               link to the system native TLS library
tls-native-vendored      automatically build a copy of OpenSSL and statically link to it

Building with the tls-insecure feature adds a --tls-insecure command line flag which disables certificate verification. Use this with extreme caution.

The tls-* flags only affect the TLS library used for the WebSocket connections (to the XMPP server and to the JVB). Gstreamer uses its own choice of TLS library for its elements. DTLS-SRTP (the media streams) is handled via GStreamer and uses automatically-generated ephemeral certificates which are authenticated over the XMPP signalling channel.

Building with the log-rtp feture adds a --log-rtp command line flag which logs information about every RTP and RTCP packet at the DEBUG level.

Debugging

It can sometimes be tricky to get GStreamer pipeline syntax and structure correct. To help with this, you can try setting the GST_DEBUG environment variable (for example, 3 is modestly verbose, while 6 produces copious per-packet output). You can also set GST_DEBUG_DUMP_DOT_DIR to the relative path to a directory (which must already exist). .dot files containing the pipeline graph will be saved to this directory, and can be converted to .png with the dot tool from GraphViz; for example dot filename.dot -Tpng > filename.png.

License

gst-meet, lib-gst-meet, nice and nice-sys are licensed under either of

• Apache License, Version 2.0, (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

The dependency xmpp-parsers is licensed under the Mozilla Public License, Version 2.0, https://www.mozilla.org/en-US/MPL/2.0/

The dependency gstreamer is licensed under the GNU Lesser General Public License, Version 2.1, https://www.gnu.org/licenses/old-licenses/lgpl-2.1.en.html.

Contribution

Any kinds of contributions are welcome as a pull request.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in these crates by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Acknowledgements

gst-meet development is sponsored by AVStack. We provide globally-distributed, scalable, managed Jitsi Meet backends.