Sound

Once the sound has started forwarding (and not before), you may get more diagnostic messages by setting XPRA_SOUND_DEBUG=1. (see Debugging) XPRA_SOUND_QUEUE_TIME can be used to control the default amount of buffering on the receiver.

Status and Platform Quircks

Sound is fairly stable in v0.9.x onwards on Posix platforms, as long as you use the latest version (see #396), there are still quirks to be ironed out for MS Windows (see #362) and OSX (see #379).
The only codec that is widely used is mp3.

Useful development documentation:

• ​queue element
• ​Pipeline manipulation
• ​Buffering
• ​Basic tutorial: Streaming

Requirements

You will need:

• a build/installation including GStreamer support for both client and server. We need the "base" plugins and the "ugly" plugins (for mp3)
• a client with sound output of any kind (even virtual)
• a server with pulseaudio installed (for going beyond the simple test step)
• users must be "pulseaudio-enabled" - whatever that means in terms of permissions and user setup (policy, user group, etc: this is distribution specific)

Standalone Tests

You can run the following test commands to run/test individual parts of the sound pipeline:

• ./xpra/sound/gstreamer_util.py (named GStreamer_Info.exe) on MS Windows) - will dump all the gstreamer plugins installed, validates that the gstreamer installation is functional. At present, you will need at least the following plugins for mp3 encoding (mp3 is the default codec):
  • for the server: pulsesrc, audioconvert, audioresample, lamemp3enc and appsink
  • for the client: appsrc, mp3parse, mad, volume, audioconvert, audioresample, queue and pulsesink
• ./xpra/sound/pulseaudio_util.py will list the pulseaudio devices found if a pulseaudio server is running
• ./xpra/sound/src.py will dump the sound coming from the pulseaudio server to a file until interrupted (named Sound_Record.exe on MS Windows)
• ./xpra/sound/sink.py will play a sound file until interrupted - a "cheap mp3 player" (Sound_Play.exe on MS Windows)

Test Sound Source Setup

The easiest way of testing that a system is capable of forwarding and receiving sound is on a Posix system (Linux, BSD, etc). Using the XPRA_SOUND_TEST environment variable, we can instruct the xpra server to use a fake sound source instead of hooking into a pulseaudio server (which we do not have to / want to start, yet):

XPRA_SOUND_TEST=1 xpra start :10 --no-pulseaudio

Then we can just attach the client from the same system:

xpra attach

You should then hear a constant (rather annoying) tone. Getting this far ensures that sound forwarding does work on this system, albeit from a fake source instead of a pulseaudio server.

Test Sound Output

Since we use ​GStreamer for sound support, the first thing to do is to check if gstreamer sound output works on your system. Using gst-launch:

gst-launch audiotestsrc ! audioconvert ! audioresample ! autoaudiosink

If the autoaudiosink does not work for you, you may want to try other options (availability varies): pulsesink (PulseAudio), osxaudiosink (OSX), directsoundsink (MS Windows), alsasink / osssink / oss4sink / jackaudiosink (Posix). If this does not work, please refer to ​Running GStreamer Applications documentation (in particular GST_DEBUG)

Simple Diagnostics

Things to look for:

• if the speaker is greyed out in the tray then sound is not enabled, check:
• the speaker and/or microphone options are enabled in the default xpra.conf file, remove any --no-speaker or --no-microphone options.
• Session Info shows sound attributes in 'Features' (see example screenshots below)
• "xpra info" should show the pipeline state when a client is playing sound:

  $ xpra info | grep -i speaker
  speaker.state=active
  

• pavucontrol both on the client and inside the xpra session (see example screenshots below)

Using an existing pulseaudio server

Next, we want to add pulseaudio, but without starting it ourselves, so we can simply re-use the existing server that gets started with most modern desktops.
Check that pulseaudio is running in your Posix desktop:

ps -ef | grep pulseaudio

Start an xpra server:

xpra start :10 --no-pulseaudio

Or create a shadow of the current desktop session:

xpra shadow :0 --no-pulseaudio

Attach from the same machine:

xpra attach

The server log file should contain the warning message:

identical pulseaudio server, refusing to create a sound loop - sound disabled

Getting this far should ensure that all the components are installed, enabled and that the xpra server found the pulseaudio device to connect to.

Test Remote Setup

Using a second client machine (or a second desktop session, virtual machine, etc), we can connect this new client to the server and get the sound from the desktop session forwarded to the client. The server can be running the test source or a the desktop session's pulseaudio server. In the case of an existing pulseaudio server, simply start any sound application in the desktop session, the sound should come out in both this session's speakers (if any) and the client's speakers (if any / different!). pavucontrol on the server should show an Xpra entry in the Recording tab. The client should show an Xpra entry in the Playback tab (or whatever mixer/sound tool you have installed) - see screenshots below.

Full Setup

Finally, we may want an xpra session to use a dedicated pulseaudio server. This is more difficult and at present, unless you really know what you are doing and are prepared to pickup the pieces, you should not attempt to do this as a user which already has an existing pulseaudio server running. Simply clearing the environment and trying to start a new pulseaudio instance will not work without some serious hacks to the environment variables that pulseaudio uses to find existing instances. If this is what you really want, ​winswitch will do this for you.
Now, assuming that you have a dedicated user, without any pulseaudio server running, and that the pulseaudio option is enabled (ie: no --no-pulseaudio command line option and no pulseaudio = no in the xpra.conf file). You should be able to setup a sound-capable environment by using simply:

xpra start :10

If, this does not work, you may want to try to start the pulseaudio server manually via:

xpra start :10 --no-pulseaudio \
  --start-child="pulseaudio --start --daemonize=false --system=false \
                --exit-idle-time=-1 -n --load=module-suspend-on-idle \
                --load=module-null-sink --load=module-native-protocol-unix \
                --log-level=4 --log-target=stderr"

Notes:

• you will need to use a tcp socket (--bind-tcp= or ssh forwarding to connect to this account - out of scope here.

The default log-level in the default xpra.conf file is low, but if you raise it to 4 or above (ie: if you use the explicit command line above to start pulseaudio), you should then see these messages in the server log:

(..)
D: [pulseaudio] main.c: Got org.PulseAudio1!
D: [pulseaudio] main.c: Got org.pulseaudio.Server!
I: [pulseaudio] main.c: Daemon startup complete.
2013-01-07 17:42:55,732 xpra is ready.

And when connecting a sound capable client:

I: [pulseaudio] client.c: Created 0 "Native client (UNIX socket client)"
(...)
2013-01-07 17:44:32,205 will send sound using mp3 codec
(...)
2013-01-07 17:44:32,309 starting sound using null.monitor
D: [pulseaudio] module-suspend-on-idle.c: Sink null becomes busy.

And yet more log messages as your start a sound application in this xpra session.

ALSA, OSS, etc

The instructions above will make sound work with all applications that use pulseaudio for output. For those that do not, please see ​PulseAudio Perfect Setup to route their sound to pulseaudio.

Screenshots

• Sound information displayed on session info (with speaker enabled and running and microphone disabled):

• A Linux client's pavucontrol showing the Xpra application connected to the local pulseaudio server:

• pavucontrol running within the xpra session ("on the server"), showing xpra recording the session's sound: