the idea is this

the werckmeister midi player uses a javscript version of fluidsynth and therefore we use soundfonts. To minimize the net traffic we compose the soundfont on the fly on client side. To do so we download the needed samples before.

sfsplit

this program splits a soundfont file into a skeleton, where all the sf header data is located, and into the sample files.

sfcompose

this program is compiled via emscripten and runs on the client. It composes under the usage of the skeleton file and the sample data a soundfont file only with the needed sample data.

workflow

for example using the FluidR3_GM sounfont

sfsplit

• create a output folder
• copy FluidR3_GM.sf2 into the folder
• execute sfsplit $out/FluidR3_GM.sf2

this will create a FluidR3_GM.sf2.skeleton file and ~1400 sample files: FluidR3_GM.sf2.<sampleid>

sfcompose

get the needed sample ids

• use sfcompose with the getsampleids command:
  • sfcompose $pathToSkeleton --getsampleids [banknr presetnr]
  • for example we want bank 0, preset 16 and bank 1, preset 5
  • sfcompose out/FluidR3_GM.sf2.skeleton --getsampleids 0 16 1 5
  • you get a list like this 0,1,2,4

use sfcompose to create the soundfont

• sfcompose $pathToSkeleton $pathToSamples $samplePathTemplate $outfile [banknr presetnr]
• the samplePathTemplate means the part of any sample file before the number. So for samples like FluidR3_GM.sf2.*.smpl the samplePathTemplate is FluidR3_GM.sf2. * for example sfcompose out/FluidR3_GM.sf2.skeleton out FluidR3_GM.sf2. mySoundfont.sf2 0 0 0 16

Sources

polyphone
MuseScore