Short audio files don't get played completely

[audiofranky]

I am creating a mapping for some drum kit library, and noticed that very short files don't get played out correctly. This happened under Linux, against JACK backend, using sfizz-git, branch "development", commit ID ce2170c.

I have attached a ZIP archive testdata.zip with an example .sfz and 3 short files of pink noise to show the effect.
The 0.2s and 0.1s duration files get played out correctly; the 0.05s file is only played partially.
This is also seen in the enclosed screenshot from a jack_capture run of that session.

Even though this might look like a "synthetic" test, there are in fact very short (hihat) samples that have roughly the same duration.
I assume this is not some kind of automatic "noise gate" kicking in?
If more info is needed, let me know. Thanks.

[paulfd]

There is some form of noise gate but it should not be active here. Thanks for the report !

[jpcima]

This is possibly another case of the short sample being interpreted as a wavetable.
Can you try with oscillator=off?

[audiofranky]

Can you try with oscillator=off?

Yes, that seems to be it! I added "oscillator=off" after the tag in several problematic .sfz drumkit files, and in all cases the swallowed or truncated samples now play correctly. So I guess it's an "intelligent" feature that one should be aware of.
The documentation is not very clear about this opcode: https://sfzformat.com/opcodes/oscillator
I wonder whether its default should be off? Anyway, thanks for the hint, most welcome!

[jpcima]

It is behavior which imitates cakewalk sfz.
There is a paragraph which mentions this.

[audiofranky]

Ok, understood..thanks for digging this out.
I still feel like there might be some more drumkit SFZ writers like myself who will find this irritating.
E.g. synthetic Hi-hat samples can be pretty short. Well, whatever, I learnt my lesson :-).

[jpcima]

In the end I'm not sure that this feature does us a great service.
It breaks expectations of a user coming from sforzando.
On the other hand sforzando has never implemented the wavetable feature. (not that I know anyway)

[AlexMitchellMus]

Then again @jpcima sfizz could still support wavetables with oscillator=off as default.

The automatic setting to interpret all short samples as wavetables feels like a drag-n-drop feature anyway. So if a user drags a short sample onto sfizz, the generated sfz could set wavetable to on verbosely.

But even then I think there are better ways to be clear as to the outcome for DnD.

I'm not the biggest fan of magic behaviour, and this has always felt like it.

[jpcima]

In cakewalk dropzone, this definitely is a drag-and-drop feature.
Most sound files which are wavetables don't contain a discriminating feature that identifies them with certainty.

But we could use the audio size in frames. Wavetables will extremely likely be a power of 2.
It reduces significantly, although not totally, the probability of error.

[paulfd]

In cakewalk dropzone, this definitely is a drag-and-drop feature.
Most sound files which are wavetables don't contain a discriminating feature that identifies them with certainty.

But we could use the audio size in frames. Wavetables will extremely likely be a power of 2.
It reduces significantly, although not totally, the probability of error.

Honestly if this is confusing where < XXX samples, I think a rule that says < XXX samples AND a power of 2 is even more confusing for non-developers.

I'd be more in favor of disabling the auto-detection entirely and let users decide to read the file as a wavetable.

[MTornPDX]

I'll put in a vote to disable the auto-detection. It caused me a lot of confusion at one point.
Using wavetables is a deliberate and specific choice, so having to enable them manually seems reasonable.

[jpcima]

I'd be more in favor of disabling the auto-detection entirely and let users decide to read the file as a wavetable.

Another possibility, it's oscillator=off as default, and oscillator=auto in case of importing audio files.
This preserves the drag-and-drop functionality.

[AlexMitchellMus]

For DnD function, why do we need auto? Could not the importer set the oscillator mode to on if the sample is heuristicly determined to be a wavetable?

[jpcima]

heuristicly determined

it's what oscillator=auto does.

• size < 3000 frames
• has wavetable chunks in the audio file