Re: [Denemo-devel] MIDI generation

[Richard Shann]

On Sun, 2012-02-12 at 18:09 +0100, Nils Gey wrote:
> Am Sun, 12 Feb 2012 16:51:14 +0000
> schrieb Richard Shann <address@hidden>:
> 
> > It occurred to me a month or so ago that with just a few primitive
> > commands added to Denemo we could generate the MIDI for playback from
> > scheme. This would make it much easier to implement things like
> > repeats, less mechanical note lengths and so on.
> > With Nils working on a more MIDI oriented program, I wondered if there
> > was a possibility of developing the algorithm for MIDI generation in
> > an abstract form, so that it could easily be implemented for any
> > program by providing primitives for the basic operations of stepping
> > through a score.
> > Does anyone fancy writing such a top-level description of MIDI
> > generation from a score? Has anyone listened to other attempts at
> > this, e.g. LilyPond's MIDI output?
> > 
> > Richard
> 
> 
> I use Python but I still use libsmf, the python bindings are done by
> Dominic Sacre.
> 
> The core of my midi generation is a cursor that goes through the
> score, step by step from the beginning to the end. Each object in the
> score has "exportMidi" method,
hmm, how do tied notes work in that case - they just export a start but
no end? With the tied-to note doing the end but no start? Yes, that
would work - harder would be the sort of tenuto which implies notes
overlapping slightly. You need to output the end note for a later time
than the start note of the next note. Will you be able to handle this?

>  which returns or directly adds the midi
> byte object to the midi stream/libsmf.
> 
> The clue here is that the cursor gathers data which can be used for
> better midi generation. For example if there is a keysig change or
> instrument change the cursor can save that and all following notes can
> rely on that data. This example could be used for different tunings and
> tuning changes, if it weren't possible anyway by just sending a tuning
> command to the sampler.
> 
> The real problems are repeats. I have not implemented them myself. A
> simple repeat maybe would be simple but once you get first and second
> volta and nested repeats it becomes more difficult.


I would think repeats would be handled by making a first pass
establishing which sections are to be played and the order they are to
be played in. Then a second pass would traverse the sections (repeatedly
as needed) obeying directives to do things like playing the repeat more
quietly.

 I imagine a good MIDI generator would also keep track of which beat in
the bar a note was on, and consult something to say how each beat should
be played, (e.g. slightly louder ...)

Well, I was just wondering if we could develop a written description of
what should be done, an algorithm, which could then be implemented for
any score-based program, rather than hard-coding the algorithm in some
code.

Richard



> 
> Nils
> 
> _______________________________________________
> Denemo-devel mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/denemo-devel