SoundFonts | Powerful polyphonic synthesizer for iOS | Audio Utils library

I'll answer my own question, but the very big part of the debug/solving process is in the question itself done with updates.

The last part was to get it to sound, when not even waiting with sleep was making it work. Not even doing it before and after the play_note function. May I say that this function always returned True, so the note was expected to sound from the very beginning. The thing is, that the SF2 file (almost 150MB) was loaded successfully, or at least it seemed like it, since it returned True as well and was for sure pretty fast.

The solution

Let's continue after reaching the point where no errors were printed when executing my script (just before Update 3 in the Question).

I wanted to check how much CPU percent usage was doing my script, so I used top on my Linux terminal and found pulseaudio running from several days ago:

Killing this process allowed it to sound finally. However, I must say that a time.sleep() with about 0.25 seconds was added after the play_Note() function, in order to let it play the note completely.

Internally, fluidsynth places all soundfonts on a stack. Because palm-muted-guitar.sf2 is the last one loaded, it is top most on the stack. When e.g. a program change event occurs on a channel, fluidsynth looks through the soundfont stack from top to bottom, searching for a soundfont that provides the requested bank/preset combination. palm-muted-guitar.sf2 is the first one to provide a percussion instrument on bank 128, preset 0, thus it is selected.

That said, your MIDI file probably sends a program or bank change event on channel 9. You may edit the MIDI file and get rid of those events. You may edit palm-muted-guitar.sf2 and remove the drum preset. Or you may setup a midi router that discards all program and bank change events on that midi channel.

player.reset-synth is irrelevant, as you are not using fluidsynth's midi player to play MIDIs.

Solved. Seems the project even though syntax highlighting all methods correctly was somehow still referencing an older AudioKit version's framework. After cleaning, emptying DerivedData and installing AK via Cocoapods as opposed to direct framework link the AppleMIDISampler now loads soundfonts as expected.

As I wrote the final paragraph, I realised:

fluid_synth_process() provides no mechanism for specifying timing information or sample offset. Yet we observe that time advances nevertheless (each block is different), so the simplest explanation is: the FluidSynth instance begins at time 0, and advances by numSamples*sampleRate seconds every time fluid_synth_process() is invoked.

This leads to the revelation: since fluid_synth_process() has side-effects upon the FluidSynth instance's timing: it is dangerous for multiple voices to run this upon the same synth instance.

I tried reducing const int numVoices = 8; to const int numVoices = 1;. So only one agent would invoke fluid_synth_process() per block.

This fixed the problem; it produced a perfect waveform, and revealed the source of the discontinuity.

So, I'm left now with a much easier problem of "what's the best way to synthesize a plurality of voices in FluidSynth". This is a much nicer problem to have. That's outside of the scope of this question, and I'll investigate it separately. Thanks for your time!

EDIT: fixed the multiple voices. I did this by making SynthesiserVoice::renderNextBlock() a no-op, and moving its fluid_synth_process() into AudioProcessor::processBlock() instead — because it should be invoked once per block (not once per voice per block).

Aaah, yes, yes, yes! I finally found a lot of things by eventually bothering to read "everything" in the doc (diagonally). And let me tell you this api is a diamond in the rough. Anyway, they actually have what I wanted with Audio param :

The AudioParam interface represents an audio-related parameter, usually a parameter of an AudioNode (such as GainNode.gain). An AudioParam can be set to a specific value or a change in value, and can be scheduled to happen at a specific time and following a specific pattern.

It has a function linearRampToValueAtTime()

And they even have an example with what I asked !