Coarse Tuning RPN should shift the key number internally

[spessasus]

FluidSynth version

FluidSynth runtime version 2.5.4
Copyright (C) 2000-2026 Peter Hanappe and others.
Distributed under the LGPL license.
SoundFont(R) is a registered trademark of Creative Technology Ltd.

Describe the bug

According to General MIDI Level 2:

3.4.3 00H / 02H Channel Coarse Tuning
...
[recommended]
This transposition function can be implemented by shifting the MIDI Note numbers internal to the synthesizer.

This is not followed by fluidsynth which artificially increases the pitch of the sample by the specified semitone amount.

Expected behavior

The keys should internally be shifted.

The C major scale in steps to reproduce should sound naturally, exactly like playing the notes normally.

Steps to reproduce

1. Load this MIDI file.
2. Load the General User GS soundfont.
3. Hear that the C major scale with a single piano sample being forcibly pitched up.

Additional context

No response

[derselbst]

Thanks for digging into these topics. I currently only have access to my mails, so I cant test this. Yet I’d like to understand this recommendation. To “shift notes” two approaches come to my mind: manipulating noteon keys or sample root keys. Messing with noteon keys seems counterintuitive, because it’s used for instrument zone selection and for matching noteoff events. Adjusting the root key seems ok for me, but how would it be different from fluidsynths current behavior? (I understand it would be different if instruments used a custom GEN_SCALE but I don’t think this is the case for your example). Oh, and what about instrument zones with fixed key generator - would they be affected as well?

[spessasus]

Yes, I meant the noteon key. And I exactly meant to change the preset zone selection because this is what GM2 recommends. You can clearly hear this with Sound Canvas VA (a GM2 certified device) with the piano and the attached test. The piano samples clearly change even though we're pressing the same key.

So essentially, we shift the key very early, just at the start.

For example:

• key shift of 2 (2 up)
• note on 60
• fluid synth treats it like 62 was passed to the argument
• looks up info for key 62, etc. So the fixed key doesn't really matter here
• note off on 60
• again, this is added with the channel key shift, resulting in a note off for 62.

I don't think it's counterintuitive and it makes perfect sense IMO. we want to change the zone selection so it plays naturally, just like SCVA does. Because currently were forcing a sample that's clearly not meant for this key to play as it and it sounds unnatural. But coarse tuning is just a simple key shift, not a glorified pitch wheel like it currently is. So I don't see any issues with this approach, since we shift note off as well.

[spessasus]

Also here's a very important test, which reveals a mistake in my previous reply:

rpn_tuning_real-time_test.zip

This tests sends pitch change BEFORE turning the note off, yet it still works in SCVA. So the note number should strictly affect the preset selection, not the MIDI note number of the voice.

Example:

• Key shift of 2
• Note on 60
• Voice has a MIDI note value of 60
• Zones are selected for key 62
• Note off 60
• Voice matches note 60, so it turns off.

I was not aware of this before, but I tested it and it's important. I think.

[derselbst]

Thanks for the extra explanation. Using RPN that way for preset zone selection and key pitch is interesting and plausible. I’m just not yet entirely convinced what to do with fixed key generators. Suppose the RPN coarse tuning causes a preset / instrument zone with a fixed key generator to be picked. From my humble understanding I would say that the voice should play at exactly that key pitch given by the fixed key generator and regardless of the RPN coarse tuning. Would you be able to test that?

…

Message ID: ***@***.***>

[spessasus]

I can test that with my own synth, but it's not GM2 certified obviously. There's no GM2 certified SF2 synth AFAIK.

Though I'm not sure how the generator matters here. We treat it exactly like we usually do.

Essentially, voice has 2 key numbers: midi note and internal note. MIDI note is directly passed from noteon, nothing changes it (not even the key generator, because that would break note offs entirety).

And then we have the internal note. It is the note used for zone lookup, and then can be potentially be overridden with the key generator. This is the note used for scale turning and all internal sample pitch calculation. And this is the note we apply the key shift to.

At least, that's how I implemented it. It seems to work flawlessly and makes sense IMO. You can try this solution on the api-changes branch of spessasynth_core to see how it behaves. Or feel free to upload a test sound font and I can test it as well.