Artifacts present in the harmonica preset of the SF3 version of the default soundfont.

• Dec 19, 2016 - 00:34
Type
Functional
Severity
S5 - Suggestion
Status
active
Regression
No
Workaround
No
Project
MuseScore
Tags
soundfont

The problem is outlined here in this support form thread....
https://musescore.org/en/node/151406#comment-617306

At first I thought it was a bug in Fluidsynth, until I realised I was (of course) playing the SF2 version in Viena.

Werner's SF3 encoder is again introducing artifacts into some sounds - I currently have a similar problem with the trumpet, which is why there is no new version of FluidR3Mono_GM.SF3 uploaded,

Initially I am going to try a recompile of the SFtools encoder, but if that doesn't work someone may have to look at the code again, which is way beyond my understanding!

Comments

• Dec 20, 2016 - 10:16

The fresh compile of SFconvert has improved the level of distortion in the harmonica, but not eliminated it completely.

This means that someone with knowledge of how the sfconvert code works needs to take a look at this.

One issue may be that the loop end points are very close to the end of the samples, which may be causing issues.

I'm therefore unassigning myself from this.

Reply

• Sep 1, 2017 - 10:05

By looking at the waveform, it seems that there are no evident problems.
I fear it is a problem strongly relate to the compression ratio. At the moment the quality factor (compression level) is hardcoded in sftools at 0.3 (from the available range -0.1 to 1.0, low to high quality).
I tried to change the quality factor to 0.6, and here is the result:
https://drive.google.com/open?id=0BxjayMZiuupOTEJ6WldadDBOMmM
from the small and quick test with the score sample I am using to check the bug I cannot hear the distortion (but my speakers are very bad and I can only faintly hear the distortion even with the original sf3).
@ChurchOrganist : could you please try it, also for the trumpet problem you were mentioning in the other thread?

Of course, increasing the quality level brings to an increase in size of the sf3 file (in this case from 13.8 MB to 18.7 MB).

Edit: I forgot to say that this new sf3 is based on FluidR3Mono_GM 2.315

Reply