KDE 3.0 Multimedia Meeting

Stefan Westerfeld has posted a summary of a IRC discussion held by the KDE Multimedia team last month. Essentially a KDE 3 roadmap for the multimedia team, the discussion covers topics ranging from MCOP and OSS compatability to recording and video embedding. A slightly edited version of his post follows.

This summary provides an overview of the things that we talked about during the KDE3.0 multimedia meeting which took place on IRC (2001-09-04).

We talked our way through more or less the topics that were posted before on the list. I will summarize this document to a list list of keywords for the KDE3 planned features document (which is relevant for the freeze), but this is the more verbose source of what will be done.

1. Making the MCOP component model more open to the world
2. Compatibility with legacy/broken OSS drivers
3. Enabling i18n
4. Recording
5. Video embedding / streaming
6. PlayObjects (special point ;)
7. MIDI
8. GSL Engine
9. GSL other stuff
10. Samples
11. Environments
12. Decisions & Open Discussion
13. Conclusion
14. Meeting Information

Making the MCOP component model more open to the world

• MCOP<->DCOP bridging: the plan is to provide access to MCOP based services from DCOP - thats useful for scripting, as well as development, and desktop integration.

==> responsible: Nikolas Zimmermann


• JMCOP: There is a working but unreleased partial Java MCOP implementation, so if you want to work on something like this, don't start from scratch - however, there is currently no fixed completion date for this, nor will it probably be distributed bundled with KDE3.0.
• UCOM: There was a bit discussion about this one, but nobody seemed to have looked too much into UCOM yet.

Back to top

Compatibility with legacy/broken OSS drivers

• Threaded OSS support: some applications acquire better results with respect to compatibility with various possible kinds of hardware than aRts, because of their threaded implementation of accessing the sound card. The threaded oss patch is a patch that opens the driver in blocking mode and spawns a seperate thread.
   
  While there was discussion whether the observed increase in quality (less dropouts, better support on more hardware) was caused
  • additional buffering gained through this (especially for cards with small hardware buffer)
  • brokenness of various OSS drivers for completely non-blocking I/O
  • the usage of select(...) which might be badly implemented for some OSS drivers or a combination of those, it generally was considered a good idea to do it
  
   
  ==> responsible: Matthias Welwarsky [ I wrote him here because he wrote the patch, but since he wasn't at the meeting, we might need to change this ]
  
  

Back to top

Enabling i18n

• All strings emitted by aRts should be internationalizable in KDE3.0 - they are currently not, because aRts does neither use libkdecore (so it can't use the standard i18n), nor has own means to deal with that.
   
  While workarounds might be possible (such as artsd only emitting error codes, and seperate in-the-gui-layer-translation), the right solution, which was also somewhat discussed on the lists will be adding own i18n capabilities to libmcop. Whether or not libc i18n can be used, or whether the kdecore code could be copied, is still to be evaluated. As there are no unicode strings in aRts (no QString), utf8 is probably the way to go for transport to the GUI.

==> responsible: Jeff Tranter [ Volunteered on the lists recently ]

Back to top

Recording

• The aRts C API didn't implement clientside recording until now, but it should. Likewise should the Arts::SoundServer interface. It might be good finally following the KDE2 plan to provide a KAudioStream class in libartskde, to provide recording/playing streams in a nice, canned, easy-to-use way. Including an end user sound recorder into kdemultimedia might be useful, too.

==> responsible: Matthias Kretz

Back to top

Video embedding / streaming

• There is an VideoPlayObject interface in KDE2.2 already, which supports playing videos via X11 embedding. This is a quite simple solution, but using it more in KDE applications will make viewing videos much more user friendly than it is now.
• There was some discussion about videos and noatun. Noatun currently doesn't use the embedding interface, and this might be hard to do, because every skin needs to be changed for that. On the other hand, noatun is the official KDE media player, so it should probably be able to embed videos.
• There was some discussion about eventually implementing video editing, but we currently lack mainly the codecs to do it. The codecs don't support precise frame based seeking and delivery of the raw image as data. Besides this, video support would greatly benefit from introducing more structured asynchronous data delivery (with timestamps and frames) into aRts.
• Extending Brahms towards embedding on the other hand seems less hard, so that it might be reasonable simple to compose film sound tracks with Brahms eventually, or - if we have better codecs - even edit films the same way Brahms can edit music.

Back to top

PlayObjects

• Malte does have a RealMedia PlayObject ready (based on the Real library, so the implementation is not entierly open source). Currently, there are unsolved licensing issues.

==> responsible: Malte Starostik
 

• There was some discussions on other ways to get codecs. Basically, writing a PlayObject for one of
  • ffmpeg
  • libmpeg3
  • xine
  • gstreamer
  could extend the fileformats aRts will be able to play. Martin Vogt is working on xine, combined with code to standarize rendering and output more, to make integration of new video codecs, and later, video editing, simpler. There is experimental libmpeg3 code in the CVS.

Back to top

MIDI

• Midi sequencers (Brahms/Rosegarden/Anthem) have started to use aRts as way of outputting midi, whereas other applications (kmid) still use libkmid, and some things (for instance ALSA) are not supported by aRts at this time. For KDE3.0, aRts should provide a complete solution.
• The midi interfaces should finally support accessing sound card hardware of all fashions (ALSA, OSS-hw-synth) - that will mean a rewrite of some stuff, but is necessary for making the midi interfaces reasonable complete.
• Precise synchronization between midi/audio should be adressed.
• Libkmid could finally be moved upon the new midi interfaces.
• The midi libs should be mainstream (i.e. kdelibs) and stable after KDE3.
• A midi PlayObject would be helpful, to enable media players like noatun to deal with midis.

==> responsible (all midi issues): Antonio Larrosa, Stefan Westerfeld

Back to top

GSL Engine

• GSL will be the new engine for scheduling synthesis modules. This will not cause a glib dependancy.
  
  ==> responsible: Tim Janik, Stefan Westerfeld
   
  
• more lengthy explanation about what/why/how --
  aRts currently has an engine that decides which modules are to be calculated when, given a certain flow graph - this is called scheduling (and implemented in kdelibs/arts/flow/synthschedule.cc)
   
  The GSL engine is a complete replacement of the algorithm with the following characteristics:
  • it should be quite a bit more efficient in the common case
  • it supports dispatching calculations in multiple threads, which for instance allows to fully exploit multi processor systems
  • it allows putting module calculations in a seperate thread from the aRts main thread, thereby avoiding dropouts from lengthy aRts operations, such as for instance calling open() on an NFS drive
  • it is designed in a way to support advanced scheduling features that aRts currently does not support, that is
  • feedback loops (circles)
  • sample precise timing
  • it should allow achieving very very low latencies reliable (where very very low is something like 3ms) - aRts currently can't do this
  • it is implemented in plain C, using some glibisms, however it can be integrated into aRts with a glib-compat source in a way that it does not bring additional dependancies (no glib required)
  The GSL engine should first act as a drop-in replacement - however to take advantage of its capabilities, it will be necessary to make sure that the modules are implemented in a way that they can run in a seperate thread without interfering with anything else.
   
  That means, that there will two types of modules: the old modules (which will still be running in the aRts main thread), and the hard-rt-modules (which will be able to run in the engine threaded).
   
  You will get the very very low latency benefit only if you use only hard- rt-modules, but everything should work with any mix of modules.
   
  Where is GSL from?
   
  GSL is designed by Tim Janik and me, and contains a lot of know how about music apps (Tim Janik has been working for years on Beast - best.gtk.org, and I have been working for years on aRts).
   
  GSL code is in the CVS and working (although yet somewhat slow) already.

Back to top

GSL other stuff

• GSL provides a sample cache which supports partial caching and preloading, the integration of that will get rid of the long outstanding issue that aRts always needs to load whole WAV files (or other formats), and support GigaSampler like "read-from-HD-and-play" style sample sets of huge size.
• GSL provides filter design code for butterworth, tschebyscheff-1/2 and soon elliptic low/high and soon bandpass/stop filters, which can be wrapped in new aRts modules (aka Arts::Synth_HIGHPASS).

Back to top

Samples

• While aRts is good at creating synthetic sounds and using these as instruments in a sequencer (i.e. Brahms), playing samples will greatly enhance the possibilities.
• I have a prototype of a .PAT loader - you can import a whole timidity style patchset with that, which means that we have a useful GM-sample-set for aRts. There is also AKAI loading code in kmusic/ksynth. GSL provides another native format. All of these probably should use the GSL sample cache for efficient, preloaded, replaying.
• Combined with an aRts PlayObject, media players (noatun, kaboodle) gain not only the capability to do similar replay than kmid (output midi via card synth or external midi), but also kmidi (output midi as calculated audio data from sample sets).

==> responsible (all points): Antonio Larrosa, Stefan Westerfeld

Back to top

Environments

• The work started with mixers/environments should be completed - for instanceit would be very useful if the mixer could save/restore itself properly, and if it could also contain insert-stereoeffects and a master volume. Also, the mixer look more like a real-mixer(tm), for instance groups or panning is not there right now.
  
  ==> responsible: Matthias Kretz
   
  
• Visualizations (led-bars) would be nice, but currently have the technical problem that they overload the MCOP layer easily, this can probably only be fixed by changing MCOP to deal more intelligently with change notifications.
  
  ==> responsible: Matthias Kretz
  

Back to top

Decisions, Open Discussion

• aRts will break binary compatibility (KDE does so, so we should, too).
• the general consensus seems to be that moving aRts in an extra CVS module (so that it is easier to obtain, compile, develop, use, package without KDE) seems to be a good thing do to
• whether to merge the SoundServerV2 interface into SoundServer (and so on), is still an open issue:
• advantages: simpler code, cleaner interfaces
• disadvantages: wire compatibility (running KDE2 and KDE3 apps on artsd) will break - or is at least very hard to keep

Back to top

Conclusion

There are about three major areas which the work focuses on. First of all, it's important to continue to improve everyday usage for casual users. Mostly, recording, i18n and the threaded OSS support fall in that category.

On the other hand, a lot of things should be simply unified, made more consistent, and merged. A media player which can play everything might sound a lot better to the user than noatun, kaboodle, kmid, kmidi and aktion.

Finally, there is a lot of work to be done especially for music applications, which includes the gsl engine, more support for samples, midi improvements, and environments/mixers.

I hope that with the way we split up the work/responsibility on several developers, we'll be able to achieve the goals listed here in time for KDE3.0.

Back to top

Meeting Information

The original IRC log can be found here. Meeting participants (in no particular order) were:

• Carsten Niehaus
• Stefan Schimanski
• Andreas Burman
• Robert Gogolok
• Achim Bohnet
• Michael Seiwert
• Stefan Westerfeld
• Matthias Kretz
• Nikolas Zimmermann
• Neil Stevens
• Tim Janik
• Jan Wuerthner
• Daniel Haas
• Malte Starostik
• Martin Vogt
• (... there was a lot of joining and leaving on the channel, so maybe I forgot somebody ...)

Cu... Stefan
--

-* Stefan Westerfeld, stefan@space.twc.de (PGP!), Hamburg/Germany
KDE Developer, project infos at http://space.twc.de/~stefan/kde *-