On 2011-05-06 19:03, Pedro Lopez-Cabanillas wrote:
Hi,
Sorry for coming so late to the discussion.
On Friday 06 May 2011, Jonathan Slenders wrote:
One thing, I moved player->user_callback(...)
before fluid_synth_handle_midi_event. Because I sometimes also wanted to
adjust the pan, the volume, and instruments on the channel played for a
note.
I can also image that someone (maybe me too) wants to alter the midi events,
supress notes or change pitch an octave up at a certain point.
Me too :-)
Seriously. This is something that I was planning for using in one of my pet projects: http://www.kde.org/applications/multimedia/kmid . This program is a MIDI/karaoke player with features like pitch transpose, tempo control, channel solo/mute, synchronized lyrics display and integrated pianola. It plays to hardware MIDI ports or soft synths. The program has three MIDI backends right now: Linux/ALSA, Windows and MacOSX/CoreMIDI. My plan was to develop a fourth backend based on FluidSynth only. It would require exactly the same feature you are trying: a callback just before delivering the MIDI event to the synth, allowing the client application to change event's parameters. But for my program this is only one of the required additions to FluidSynth. Another one would be a callback while FS is parsing the MIDI file, reporting to the client application all the meta-events like titles, track names, copyright notices, texts and lyrics, time/key signatures, that are ignored right
no
w in FluidSynth but would be required by kmid. Talking about text and lyrics,
these events would be reported by the player callback at playing time as well,
although they would be ignored by the synthesizer.
Considering the two proposed callbacks, instead of new_fluid_player2() perhaps
I would prefer two functions: fluid_player_register_playback_callback() and
fluid_player_register_parser_callback() with their corresponding unregister()
companions.
The parser callback would be quite innocuous in process time constraints, and
will be called in the same thread as fluid_player_add(), so there is little
concern about it. The playback callback is another issue: first, it will be
potentially called from a realtime thread, with small time constraints. If the
callback allows event modifications it should be called synchronously, ahead of
the delivering time to the synth, and the client application would need to
return within this time frame. It would be easier to implement a pure
notification callback, called asynchronously from FluidSynth but without
allowing event modifications.
Talking about events. The structure fluid_midi_event_t has been opaque for a
long time in FluidSynth, and I think it should remain opaque. There are API
functions to access some event properties, and more API functions would be
welcome if needed.
Regards,
Pedro
Hi folks and thanks for the patches and comments,
I mostly agree with Pedro (e g about letting the fluid_midi_event_t
remain hidden), but I'd like the callback function to be used instead of
(rather than before or after) calling fluid_synth_handle_midi_event.
That means that we'll start off with the player's callback function
pointer being fluid_synth_handle_midi_event (for simplicity and
backwards compatibility) but we can change it through a new api function
called fluid_player_set_playback_callback(). That way you should be able
to not only insert your own processing (eavesdropping, changing and
removing notes as necessary), but also to insert our existing midi
router into the playback chain for midi files.
Does that make sense or am I missing something?
(Btw - I'm holding a speech on FluidSynth real-time and thread safety
challenges tomorrow morning, if you like you can listen and ask
questions on IRC in real-time. [1])
// David
[1] http://lac.linuxaudio.org/2011/?page=program&mode=table&day=3
_______________________________________________
fluid-dev mailing list
fluid-dev@nongnu.org
https://lists.nongnu.org/mailman/listinfo/fluid-dev
