Soundfont has brighter tone than expected

• Mar 23, 2018 - 16:44
Reported version
S4 - Minor

2.2-dev. revision: 1dd229330

The sound of the "Nylon String Guitar" in this nightly is brighter (more treble) than in MS 2.1 even when using the (old) FluidR3Mono_GM soundfont. Other instruments (in the same FluidR3 soundfont) also sound brighter. e.g.



"The sound of the nylon guitar in this nightly is brighter (more treble) than in MS 2.1"

Indeed, I noticed that too. Not especially with this commit, but since the 2.2 R.C. already (and the new MS_General.sf3)

Not a good news, since the sound with FluidR3 was already aggressive ("pianistic" sound) especially in the treble notes, let say more or less from the E first string.
It's a shame, hope for an improvement in a future version.

So, what's happening here is that in MuseScore 2.2, the SoundFont is actually being rendered correctly. In previous MuseScore versions, there was an always-active velocity-based lowpass filter that was incorrectly implemented. This made every instrument sound filtered at lower velocities, regardless of the SoundFont designer's intention. Now, if we want a mellower sound at lower velocities, it must be specified by the SoundFont designer. I have implemented such filtering for the Nylon String guitar in MS_General, but it is not as extreme as what the default filter was doing previously. If people agree that it should be filtered more, I can make that change for the next version of the SoundFont.

I tried to implement velocity-based filtering in a realistic way in the new SoundFont, but if there are cases where an instrument still sounds too bright, let me know and I'll have a look at it. I want to make sure that PP passages aren't overly filtered, so I'll do my best to strike a balance. Eventually, most of these instruments will be replaced with better samples, which will be the proper solution to the cheap-sounding samples.

In reply to by s.chriscollins

"I have implemented such filtering for the Nylon String guitar in MS_General, but it is not as extreme as what the default filter was doing previously. If people agree that it should be filtered more, I can make that change for the next version of the SoundFont."

One vote here (and an affirmed vote!) to make this change and to correct, first, this excess of brighter tone for the next version.

From the first string, and the more you go in the treble, and the more it becomes aggressive to listen. From the E4 first string / A4, it starts to be bad, and higher, from the D4/E5, it's really hard, sharp, which breaks your ears, sorry :)

"Eventually, most of these instruments will be replaced with better samples, which will be the proper solution to the cheap-sounding samples."

In a second time indeed, this could / should be certainly considered.

The whole high register (even medium) in particular is not realistic at all, does not represent the sound of a nylon guitar. It is a deficiency with which one learns to live with, I would say almost alas. So it would not be luxury to try something else to get a little closer to the sound of a guitar.

I admit that I do not believe much in that. I even almost dropped my arms, tired of war. I have tried acoustic nylon guitar soundfonts, and they have always disappointed me greatly: they were definitively far away from the sound of this instrument.
Without asking for the moon and the impossible, try to do your best to get a better result.
Thanks in advance :)

In reply to by cadiz1

I made a few more tweaks to the filters on the nylon guitar, especially on the higher notes. Keep in mind that the brightness is also affected by the dynamic level used in your score. If you write FF, you will still get a bright sound. This change will be present in MuseScore_General 0.1.1, which will be released today or tomorrow.

In reply to by s.chriscollins

Yes, I do prefer the darker sound of instruments in the older versions and think that many of the instrumentinstruments (especially saw synthesizer w/ lead bass sound) is too bright or squirrel sounding for a pleasant listening experience. I agree it should be filtered more... either that or make more instruments that would sound like the older instruments used in the previous a way to reduce overall brightness using the synthesizer settings?

In reply to by Adria Sorensen

I realize that the result of this fix sounds different on everyone's hardware. Many people have found it to be a welcome change. Perhaps turning down the treble in your PC's sound settings or using different headphones/speakers might help? We can't restore the previous behavior, because it is an incorrect implementation of the SoundFont specification and would mess with being able to correctly design future improvements to the instrument sounds.

If there is enough consensus on certain instruments being too bright, then I can have a look at it, as changes in this area would have to be made per-instrument. The problem is that these things are a bit subjective, and I want to avoid the problem where making an instrument more mellow might make others unhappy.

It is interesting that you mention the new synth sounds, because the waveforms used for those sounds are rendered exactly as they would sound coming out of a real, analog synth. Perhaps you would find the real thing too bright as well?

Mike, can you elaborate? The difference between 1.x and 2.0 was the soundfont, not really anything else about the synthesis, and it was a huge change. There was also a soundfont change between 2.1 and 2.2, but it was much smaller, affecting only a subset of the sounds. If you don't like the handful of changes in the soundfont between 2.1 and 2.2, you are of course welcome to keep using the 2.1 soundfont.

But - unlike the case going from 1.x to 2.0 - there were indeed other changes to the synthesis between 2.1 and 2.2. As Chris explained, these were fixes to bugs that were causing soundfonts to not sound the way they were designed to. And as far as I can tell, that's the subject of this thread - some people who had gotten used to the incorrect sound of certain soundfonts are finding they don't like them as much now that they are finally hearing them correctly. That's too bad for the soundfont, but it doesn't really seem to be a criticism of MuseScore if it is now finally playing the soundfont correctly.

And in any case, it seems to me that if you are finding you don't like a soundfont that you previously liked now that you are finally hearing it correctly, it standard to reason you might discover you now like one you previously didn't. In any case, it seems better to simply find a soundfont you like the real sound of and stick with versions of the program that are able to render soundfonts correctly, as opposed to finding a soundfont you actually don't like but playing it through a flawed version of the program that only just happens to make it sound good.

So can you explain what specifically you aren't liking about the corrected playback, and which soundfont you find you now don't like as much as you did before? Is it only FludiR3 you don't like as much, or do you have some other favorite soundfont that you feel was adversely affected by the change? Or is it really just the handful of sounds changed in the soundfont itself? It's important to be specific in trying understand problems.

In reply to by Marc Sabatella

@S.Chris Collins and @Marc,

I uploaded months ago, when still used the 2.1 FluidR3 Mono as the default for the uploaded scores, a piece, Elegy,

I just uploaded, using my MS 2.2.1 and the new MusesScore_General default soundfont, a copy of that score. As Thomas explained weeks before, now the default is this new, tweaked soundfont, MuseScore General.

In this case, the sound of MusesScore General is good - except the Cello. I don't like the sound of the new cello, which Chris replaced at MuseScore_General 0.1 (prerelease-alhpa)

If you are interested, could you listen to both versions (below, the scores for the cello part, using 2.1 FluidR3 and MuseScore_General)? which one you like the most? I find the newer Cello has a dull or muted , or "nasal", and inexpresive sound quality. Of course, I tried to set the synthetiser (Zita) settings exactly the same as when I uploaded the score originally, but it's about the same (some slight change regarding the reverb (my default and preferred setting is having the Zita's "Output" knob to the middle, instead of the new default (one point to the left). (FluidR3 2.1)
and (MuseScore_General 2.2.1)

So, I would be interested to know your opinion about which sounds better.

To pick example bars, in bars 6-7 and 14-15 you can hear a noticeable difference in expresiveness, that with the new soundfont is lost (in my opinion of course). Another example: at bar 6 the vibrato seems to be supressed in the Ab2 eight preceding the dotted quarter (in fact, the same problem appears too in bars 14-15)

FWIW, I too generally prefer the FluidR3 cello over the new one. That's comparing both using 2.2. There are subtle differences between 2.1 and 2.2 even using the same soundfont, but I have no real preference.

In other words, for me, this does not provide evidence of a problem with the synthesis engine - that is, it is not a basis for saying that 2.2's playback is inherently worse than 2.1's. It's just a comparison of soundfonts - we're saying we find MuseScore_General's cello to be inferior (in this context at least) to FluidR3's. And it's important to keep track of which complaints are which. My guess is that the Fluid cello's relatively slower attack made it problematic in faster moving passages and that might be why Chris replaced it, but he can speak to that himself.

In reply to by s.chriscollins

OK, Chris, thanks for considering, well, every ear is different, I suppose the ideal would be to get a consensus about this.

On a related aspect (soundfont-related) I did noted you supressed a bit of the release on the Strings instrument in the new some of my scores do play not as "slurred" as they used to be :-)

In reply to by mdi1972

Yeah, it's impossible to have one setting that is perfect for every scenario, but the strings release was unnaturally long, which was particularly noticeable on separated or staccato notes. Once I am done with the new marching percussion sounds (my current, first priority task), I will finish my work on the new section strings.