The FluidR3Mono Soundfont

• Dec 15, 2014 - 18:54

As you know this has become the standard soundfont used for the 2.0 Betas.

I thought, therefore, that a link here to the latest version with all fixes would be good here.

Unfortunately my compile of Werner's compression tool is currently producing a corrupt file, so this is a link to the latest version of the SF2 file at my Dropbox account.

I shall lose some file space when my trial period ends in February, so I may have to switch to a Google Drive later.

Whatever, a link to the latest version of the .SF2 file will always be available from this thread.

https://www.dropbox.com/s/m5iv56fzbyu2hrj/FluidR3Mono_GM2-9.zip?dl=0


Comments

I have raised the volume of the drumkits in response to some complaints that they are too quiet in the forums.

So here is the link to FluidR3Mono_GM2-10

https://www.dropbox.com/s/r7c4faqs0ibv7xq/FluidR3Mono_GM2-10.zip?dl=0

I have raised the sound levels of the drumkits en bloc so you must let me know if you think some of the other drum sounds need attenuating. If all is well I shall squash it down with Werner's compression tool for use with the release version of MuseScore 2.

Hi Michael

Some observations when I tested revision 10 in MuseScore 1.3:

Volume is different between instruments compared to the original SF2 (I may have said previously, but if deliberate, fine).
In some drum kits, bass drums seems overpowering compared to snare.
Bass and snare drums of the electronic kit are all left-panned.
Fantasia sounds different compared to the original SF2.
I can still hear clicking in some instruments (claves).

In reply to by chen lung

OK thanks Chen will look at the electronic kits.

I have changed the overall volume of some of the instruments due to complaints that they were too quiet. The problem with this is that you then have to go through the entire soundfont changing balances! Not sure we have it right yet :)

I have raised the volume level of the entire drumkits so can't understand why the bass drum should overpower the snare unless it wasn't properly balanced in the first place.

The change to Fantasia could be a result of the changes to Goblin Synth I had to make to prevent polyphony overflow - it uses some of the same sounds.

There should be no clicking at all - that was a problem with the original SF3 compression due to a fault in libsndfile. I will check that.

In reply to by ChurchOrganist

I never personally had an issue with overall drumset level, but that's easy enough to tweak in the mixer. I do think the bass drum was and remains too quiet relative to snare, but I also know that my computer speakers lie to me as they reproduce high frequency sounds much better than low, and suspect the same is true for most other people. When I listen through headphones, I hear a better balance, but still a bit low in the bass drum - or maybe high in snare, or maybe both.

In reply to by chen lung

There is definitely no clicking in the soundfont itself.

If you were testing on 1.3 then there is a known issue of the Fluidsynth noise gate closing with an audible click on some sounds, notably short percussion sounds.

You can also hear this behaviour with other sounds which have a rest after a note in 1.3.

This is probably what you heard.

Michael, I'm using the soundfont included in the latest nightly (a27e320 - Mac 10.7.5), and I don't like the volume between the drums - in Standard, the snare is too quiet compared to hi-hat, for example (please check in full).

In reply to by chen lung

The hihats and snare sound in your wav are perfectly well balanced on my system.

Laptop speakers are not usually big enough to faithfully reproduce the lower frequency elements of a waveform, and so are inclined to emphasise high frequency waveforms like hihats.

I use a set of Logitech speakers with a small subwoofer unit. This is not as good quality as the reference sound system you have in studios, but is still able to reproduce bass frequencies down to a pure sine at around 30Hz.

Even budget quality headphones are much better at reproducing low end frequencies, so when checking sounds it is best to use headphones rather than laptop speakers.

In reply to by chen lung

Sounds good to me even through my laptop speakers, although I guess they are considered better than average (still pretty tinny). Definitely sounds good through headphones to me.

But of course, some day, a mixer for the individual drums in a set so users can set their own preferences would be a nice thing.

Sorry, I should have also said - even with audio-technica headphones (ATH-M35), the snare still sounds underwhelming to me. I prefer how it sounds in beta 2 (attached).

Since MuseScore is a notation program, I'm not really in favour of having controls for each drum. :)

Attachment Size
Drums [2.0 Beta 2].zip 449.82 KB

In reply to by chen lung

I don't understand.....

There have been no changes to the drumset balance since I raised the overall level of the drumsets in January.

In the clip you supply the snare is way way overpowering the hihats and the bass drum.

This makes me wonder if the default drumkit velocities have been changed?

In reply to by chen lung

I can say that your original file sounded perfectly well balanced to me. The second sounds ludicrous. I mean, it sounds fine as an example of how someone might play a rock beat where they are *deliberately* hitting the snare drum a lot harder than anything else, but that's very clearly not a normal balance between the drum.

The snare drum is clearly being hit a *ton* louder than anything else. To get that effect, you should need to use an accent mark, marking an sFz as well.

In reply to by chen lung

Are we talking about the same files? The one labelled "Unbalanced drums" sounds normal to me - the way I'd expect a drum set to balance if one were *not* deliberately bashing the snare as hard as one can with both hands at once using lead sticks.

The one labelled "Drums [2.0 Beta 2] sounds like the way I'd expect a drum to balance if one *were* deliberately bashing the snare as hard as one can with both using lead sticks.

I think you've been listening to too much rock & roll if you think that's the normal level level of a snare drum in a drum kit :-)

Load it into Audacity and you'll *see* how crazy loud the snare drum in the "Drums [2.0 Beta 2]" is in comparison to everything else.

In reply to by Marc Sabatella

I think you're both right, and you're both wrong—the truth lies somewhere in between. Absolutely, in "Drums [2.0 Beta 2]" the snare drum is way too loud. Marc Sabatella is right about that. But chen lung is definitely right that in "Unbalanced drums", as he originally said, "the snare is too quiet compared to hi-hat". I think not only does the truth lie somewhere in between, but the ideal volume for that snare sound lies somewhere in between. ;-)

The crash cymbal sample sounds like it's within the range of possibilities for a drumset, but it sounds nothing at all like the cymbals in an orchestra or concert band. Could you replace just that one specific sample with something that works for either?

In reply to by ChurchOrganist

I'm not sure how to get to the Orchestral Drumkit, but I did try adding a plain Cymbal part. It does sound slightly different from the "Crash 1" in the full drumset, but still not like a real pair of cymbals. (Also, much too quiet.)

I know people occasionally ask you to borrow a few samples from some other SoundFont. I won't ask you to do that. But I will ask you to just listen to the "Crash 1" from GeneralUser GS. Just create a drumset score, put in a row of crashes, and play it back with first your SoundFont at the top in the Synthesizer, and then GeneralUser. See what I'm driving at?

By the way, I hope I don't sound ungrateful—thank you tremendously for what you do.

In reply to by Isaac Weiss

You get to the Orchestral Drumkit from the mixer.

The default for all the drum sounds at present is the Standard Drumkit.

If you click the tab in the mixer in the same way you would for a normal instrument you will see a list of the available drumkits appear.

Scroll down till you see Orchestral Kit and then select it.

To my ears it is far more like a proper pair of orchestral/marching cymbals.

HTH

PS IMO it sounds better than the GeneralUserGS one

In reply to by ChurchOrganist

Well, in the Orchestral Drumkit the crash sounds like a kettle drum. Yes, seriously. This applies no matter what the SoundFont, and I'm about to go file a bug report.

However, one of the notes identified as "Ride (Bell)" seems to be associated with the cymbal sound, and it is quite good. I actually still like the Standard Drumkit Crash 1 from GeneralUser best, but it's difficult to say why—maybe your Orchestral Drumkit cymbal sound stays too loud too long? Nonetheless, I can't say it's not a perfectly useable cymbal sound, and I withdraw my request for you to change it. Keep up the good work!

In reply to by Isaac Weiss

I don't know that this is really possible - the same sound doens't work for both.

Defining a drum set is not overly difficult - start with the Edit Drumset button in the drum toolbar (appear while editing a drum staff) and go from there. You define which displayed notes you want to map to which MIDI sounds.

It's unfortunate that it's a biut of a pain, but we dind't get to design General MIDI. We just have to live with it.

In reply to by Jojo-Schmitz

Awesome. I'll include it in the source but unfortunately it's not easy to get to the DRM files currently.

We don't have any "orchestral" templates with percussion but we had we could also include this drumset.

Also in the future we could define the GM2 drumset and add a combo to switch to each of them.

In reply to by ChurchOrganist

Thank you! I'm trying to figure out exactly how this works. One thing I'm having issues with is that the notes seem to all be on Voice 1 in this drumset, even though according to "Edit Drumset" the bass drum is assigned to Voice 2. The notehead colors in the drum palette seem to keep changing—sometimes all blue, sometimes one green, sometimes more than one green. It's no real problem to manually change the voice of a note, but I'm guessing I'm missing something here again.

Also, if this is to be included with MuseScore, you'll want to have very clear documentation—this drumset is only meant to be used when Orchesral Drumkit is selected in the Mixer, and so forth.

In reply to by Jojo-Schmitz

Version 2.104 of the FluidR3Mono soundfont is now uploading to my dropbox account.

I've tweaked the tuning on one section of the violin, which will hopefully improve the tuning issues reported here http://musescore.org/en/node/50431

There is so much vibrato on the original samples, however, that it is almost impossible to tune them properly. I'm thinking in terms of remaking the violin instrument with fresh samples. That will have to wait until after I have finished the clarinet though.

Should be available for download in around 30 mins.

Download link https://www.dropbox.com/s/soi5tj3s3113f26/FluidR3Mono_GM_sf2.zip?dl=0

FluidR3Mono_GM.SF3 Intonation problem
While doing an arrangement for band, I'm having the alto and tenor saxes doing slow unison's and harmonies. I'm noticing that the intonation is off. The alto seems flatter than the tenor creating a somewhat disturbing playback. Anyone else notice this?

In reply to by Ashwin Batish

No, the .SF3 format was invented by Werner and is currently exclusively for MuseScore.

The latest .SF2 version is always available from my Dropbox account - the link is further up the page.

If we are going to have several people working on it, we will need to establish a version control system so that we aren't conflicting with each other.

I will think on how we can achieve this. But in the meantime feel free to experiment - these things are the epitome of the maxim "Many hands make light work".

In reply to by Ashwin Batish

To me, it's details like this that contribute to the realism of FluidR3. I would hope than in an effort to "clean up" the sample, we don't de-humanize the overall effect to the point that it loses much of what to me sets it above so many others. I guess maybe different people value different things in a soundfont.

In reply to by Marc Sabatella

No argument there Marc. It's just that what is a beautiful ornamentation and approach for one style might not sound cool in another. So we are limited by the very fact that we have to cater to a vast range of applications and the safest approach seems to be a fast attack to correct pitch. Manipulations can then be applied via a synth's capabilities of producing vibrato and protamento etc. Many years back, I was experimenting with creating vocal sound samples and came across this same problem. On one hand it sounds cool when the vocalist (in this case my father) would put the nuances necessary for an emotion filled rendition but when I sampled them, the music could only be played for that one style and any kind of speed up was virtually impossible. I never released these samples because of this reason. Ideally, a font library with as many variations is really the way to go. Great program and sounds though. A++ for all the effort from you guys here :)

In reply to by Marc Sabatella

This is why I suspect that new samples would be better than trying to clean up the existing ones.

I will leave you to judge when the replacement clarinet is ready.

Unfortunately MuseScore still lacks the desired degree of control required to produce lifelike performances.

But if we can provide better MIDI controller support in the next major version that will be a big step forward towards achieving that.

In reply to by Marc Sabatella

The thing is, "in tune" is itself a subjective thing - does it mean equal temperament for all instruments, even ones whose physical design is directly opposed to that? Show me a trumpet that produces equal tempered fifths and I'll show you one that somehow defies the laws of physics! And does "playing in tune" leave no room for the natural variations in pitch that actual human players produce - both in terms of vibrato and the natural variations that often occur on attack for wind instruments?

The goal of a perfectly controlled, perfectly in tune soundfont is often at odds with the goal of sounding *realistic*, because real human players are not perfectly controlled or perfectly in tune. Figuring out how to balance these is an art. I'm not saying I have any particualr skill in that art. I'm just expressing concern that relates not just to this but comments elsewhere about about preferring the "colorless" and entirely unrealsitc sound of some other soundfont, and would hate to see us move in that direction for our default soundfont. If you want pefectly in tune sine waves, I'm sure there are soudfonts out there for you, but I think many people value *realism*, and all the human quirks this entails. Smooth out rough edges, sure, but let's not lose sight of the big picture.

In reply to by Marc Sabatella

But the fact is that real human players—at least, the good ones—can make themselves sound a lot better in tune than this. I confess I don't know how brass players do it, but that's beside the point—I'm not talking about tempered vs. just intonation here. I'm not arguing against overtones, or vibrato, or "quirks," per se. I'm simply stating that I'd rather the alto saxophones didn't play flat.

However, you raise an interesting point with "let's not lose sight of the big picture." I actually feel like I'm the one thinking about the big picture—as in, the many instruments playing as an ensemble, and not each one focused on doing its own thing, ignoring the others, and ignoring the conductor's desperate attempts to communicate that you've got to bring that pitch up—but there is a difference between the way, for instance, a solo violin plays, and the way one in an orchestra plays, both in terms of how one treats the intervals and many stylistic choices. It occurs to me that, if you come from a perspective of writing for a single alto saxophone, you might have no problems with the tuning as it is. Quite likely, each individual instrument sounds fine on its own. The thing is, I primarily want MuseScore's FluidR3 to be a good ensemble SoundFont—and it's not there yet.

In reply to by Isaac Weiss

I appreciate your comments, but actually, I'd say you have quite the opposite impression as me. I feel many of the suggestions for improving the soundfont have been *too* based on listening to instruments in isolation and not enough on listening to the ensemble. But as I said before, different people can obviously have different perspectives. What one person perceives as "highly colored" in an objectionable way, another perceives as "realistc" in an agreeable way. What one person perceives as "more in tune" another perceives as "mechanical". What one person perceives as "too much vibrato", enough perceives as "human". And so on.

I'm not saying there is no room for improvement, but just listening to a few of my scores for small to large ensembles using the the sounfont from Finale Notepad that you've mentioned as finding better than FluidR3 - it's clearly we just have different perspectives, since I find it very significantly worse. And I suspect there would be similar disagreements on various points between any other two people participating in these discussions. That's fine, normal, and unavoidable, I suppose.

In reply to by Marc Sabatella

Make no mistake, though, I mostly like the sound of FluidR3 better than that of synthgms—it's really only the issues of these certain violin or alto sax notes throwing things out of whack that I don't like. (Though I think there are other places, I haven't made the effort to isolate them.) My ideal SoundFont would be more like FluidR3—in fact, I'd say my ideal SoundFont would be FluidR3, tuned up a little bit.

In reply to by Isaac Weiss

Totally agree with Zack on this. My score sample sounds just plain bad in spots. It's not just on the sweep up but after the pitch reaches it's intended frequency. I have a frequency reader that shows many notes are just flat enough to matter. But .... and as Zack pointed out, I love the FluidR3's sound, attack etc. Much punchier than the Sibelius or Finale. I took an xml of my score and played it in both and it was nowhere close to what the Fluid was giving back. Sib/Fin came out sounding too delicate ;) FluidR3 sounded like I had the Tower of Power working for me! The Alto could use the help and I think would actually sound great. My tests show a 3 - 8 cent difference enough for it to stand out especially when doing unison long notes. That was the point I was trying to make.

In reply to by ChurchOrganist

Thanks!
Just for good measure: do Voice Ohhs have the same issue?
They are still too quiet in the lower ranges of that new version (but better than befor), and sound pretty horrible elsewhere... May not be worth any efford though...

Hmm, not quite there with Choir Aahs, I think, listen to the attached score, and watch the meter of the synthesizer.

Some bass notes are a bit to quiet (E2-F#2, G3-A3) then C#4-D#4 quite a bit louder (too quiet before, now too loud), then G4-A4 quieter, C#5 too?

Attachment Size
SATB.mscz 8.38 KB

In reply to by lasconic

Version 2.107 is still the definitive version.

The changes to the Aah Choir in 2.108 have not solved the situation of imbalance in the samples. I am currently contemplating using samples from Mattias Westlund's Sonatina Symphonic Orchestra to redo that section, which, whilst not sounding like a genuine choir, at least are of a better quality than the current ones in Fluid.

In reply to by lasconic

I was trying to compile sftools under Mac to test if I can reproduce the sf3 corruption, but the compilation is failing with the error 'sndfile.h' not found. This is strange since I can compile MuseScore without problems (and it requires sndfile.h).
Do you perform additional stages other than cmake CMakelists.txt && make for the compilation of sftools under Mac?

I tried also the "-G Xcode" and xcodebuild compilation, but I am still getting the same error.

All done.

FluidR3Mono_GM 2.200 containing a 5 note Temple Blocks preset now uploaded to my dropbox account.

https://www.dropbox.com/s/soi5tj3s3113f26/FluidR3Mono_GM_sf2.zip?dl=0

The zip archive also contains a suggested .drm file for use with the new Temple Blocks Preset.

This is designed for a 2 line percussion staff.

To use I suggest you add the existing Temple Blocks instrument from the Instruments dialogue then go into the mixer, uncheck the Drumset Box and then choose Temple Blocks from the Patch List which you will find immediately after Gunshot in the list.

Now customise the staff to two line, enter Note Entry mode, press Edit Drumset and then load the supplied .drm file.

You can now enter the 5 blocks with the keyboard letters A to E

Enjoy.

PS Let me know if you need differently tuned sets of 5 and I will see what can be done to achieve this with time stretching the samples.

PPS SF3 version to follow once we'vee established it's working OK

Version 2.203 is now released in sf2 version which imroves some problems with the timpani.

The Timpani are still not wholly successful largely due to the instrument comprising only 3 samples all of which are very close together.

I am working on eventual replacements.

Link to the latest version here.....

https://musescore.org/en/node/94611

In reply to by ChurchOrganist

I've made a few adjustments to the new 2.312 version of FluidR3Mono SF2 (using Viena):

1. Assigned correct lower (pitch) range limit to all orchestral instruments.
2. Add 0.025 sec global vol. env. release to clarinet to remove clicking.
3. Shorter vol. env. release on harp (8 secs down to 3).
4. Added various vol. env. releases on some drumkit and orchestral percussion instruments to prevent playback being cut-off. Short release (0.025 s) to snare rolls.
5. Assigned exclusive group numbers to unassigned hats and triangles (in orch. percussion).

The amended soundfont is at: https://drive.google.com/open?id=0Bxj02NhJ9Gc_dGtDWTB1bkRWLUk (but only for the next day or so).

It's not for general download as it hasn't been checked and approved by ChurchOrganist yet. Hopefully some of the alterations, at least, may prove to be useful.

In reply to by geetar

Thank you for your contribution!

1 It's not a good idea to limit the ranges of the presets, as many if them are used for more than one instrument, and also users may use them to represent voices. Limiting the ranges in this way may well nerf pieces already on MuseScore.com.

2-3 Envelope changes were something on my list of things todo, so tyvm for saving me the job!

5 I'm not sure what this means? When I looked at the Orchestral Drumkit there seemed to be no difference between 2.312 and your changed version?

Incidentally, if you want to continue to help with soundfont maintenance versioning is now enabled via GitHub LFS. You will need to fork my FluidR3Mono_GM repository and then install Git LFS and Desktop GitHub

My repository.....
https://github.com/ChurchOrganist/FluidR3Mono_GM

Desktop GitHub
https://desktop.github.com/

Git LFS
https://git-lfs.github.com/

SF2 Version 2.313 which remaps the Marching Bass Samples to fit on a 61 note keyboard and adds the Marching Bass Simultaneous Hit is now up on Google Drive.

Please Note: The Marching Bass implementation in MuseScore 2.0.3 will not work with this version. A DRM file with the new keymappings is included in the Zip file. You will need to save this to your Styles folder and then load it using the Edit Drumset dialogue.

A new SF3 version will be available soon.

A late comment:

I appreciate the efforts made towards improving the sound (e.g. the popping issue), but I've never really been 100% satisfied with it (I'm not an expert, so bare with me) - I'll talk about each matter separately.

About drums, it could be the balance between them. For this comparison, I’ll use Standard:

I'm not totally sure yet between the skinned drums (e.g. bass, snare), but the cymbal and hi-hat seem louder (too prominent). I don't think I'm the only one (link ).

To contextualise, I produce band scores; they tend to consist of drumkits on one stave.

I attach the following:
mscz: produced in 2.0.3.1
WAV: exported from MuseScore - with and without Zita1
AIF: produced by exporting a MIDI (also attached), opening in QuickTime Pro 7.6.6 and exporting the AIF

Attachment Size
Drums.mscz 4.32 KB
Drums - MuseScore Nightly Build.mid 229 bytes
Drums (Audio).zip 1.9 MB

In reply to by chen lung

Is that the same soundfont? I'm assuming not.

The issue to me is that different genres have different expectations here. Jazz is normally more reliant on ride and hat than rock. So when you say "band" I assume you mean rock. We do have the ability to provide different kids. Maybe we should take advantage of that more.

In reply to by Arianna2001

My current plan for MuseScore 2.1 is to use ChurchOrganist's soundfont version 2.312 in order to avoid https://musescore.org/en/node/165281#comment-641176

It would be good to have some feedback on Arianna's version, so we could assess it for the next version. I would really like a community effort to make a definitive sfz orchestra for MuseScore 3. ChurchOrganist is busy but his repository at github was a good start.

It may be naive but the way I would do it is as follow:

1. List the instruments from FluidR3 and convert it to SFZ.
2. For each instrument, assess its pro and cons. Pro could be "not enough samples", "bad quality", "balancing", "wrong loop" you name it. Quality could be "good for the size", "best sound compared to other soundfont" etc...
3. For each instrument, find alternative high quality sample if necessary
4. Once we decided on the best samples. Define loop, write or get sfz file. Tweak them for best result.

In reply to by lasconic

Might be naive, but I suffer the same naivete then :-). I know one potential shortfall with this approach so that if sample come from widely different source, they might sound good individually but not together due to difference in recording techniques etc. I don't know that this is really a big concern in practice though.

BTW, for now, without knowing more about which specific sounds have been changed in which version of the soundfont, I don't think it is even worth asking which version of the soundfont to use - we should continue to use the latest version one we have, which is to say Michael's version. But we should indeed be applying the process above to compare that and Arianna's to see which of the two sounds best instrument by instrument. And we should really have a lsit of which instruments have modified and what way to help guide us in making the selections.

In reply to by ChurchOrganist

Ah. I wonder if there is any way to convince the "powers that be" regarding this format to consider extensions? Or to develop one ourselves? Or develop our own SFZ manager that knows the mapping from GM sounds to SFZ files (for this particular set of sounds, anyhow) and loads them on an as-needed basis? That's probably the right solution.

Michael,

When using the Polysynth option, there is a scratching noise (like a needle on audio vinyl) after a note has sounded.

Can you advise?

MuseScore 2.1 Nightly Build 8526f62 and FluidR3Mono_GM-sf3 2.312 - Mac 10.11.6.

FluidR3Mono_GM Version 2.315 is now available from Google Drive.......
https://drive.google.com/file/d/0B7cZM0RQwkwSZVJPQlUyTkRwY0E/view?usp=s…

Currently only the SF2 version has been updated.

This fixess keymapping problems with the Marching Tenors.

The Marching Percussion section now has two sets of Tenor Drums - one with the old mapping and one with the new, and defaults to the old mapping.

There are also now two sets of Bass Drums - the original key mapping, and a new key mapping which also includes a simultaneous hit.

Anonymous
Jun 13, 2017 - 17:54

I added the new drums to the synth and put them in the mixer to do an A/B comparison. MS prompt to put in directory, add to top of Fluid list etc. was the usual. Levels the same But both new drums didn't sound.

Attachment Size
3.jpg 114.48 KB

The ICMT chunk contains “DO NOT REDISTRIBUTE ANY OF THESE SAMPLES” which kinda is the contrary of the MIT licence. Would you remove this remark from it?

I’m wondering if we shouldn’t add the whole licence to the ICOP chunk (and then embed its content into the generated WAV, MP3, etc. files, like we do with non-sound fonts and PDF files)…

Update: I made this into a feature request “idea”: https://musescore.org/en/node/270099

Turns out ICOP is too small (255 ASCII bytes, at most), but ICMT is large enough (65535 bytes).

Do you still have an unanswered question? Please log in first to post your question.