SoundFonts Issue 1 & question 2

• Mar 22, 2009 - 20:33

First, with the workaround via Parts (see my other posts) I was able to get all instruments on my test score with 5 instruments.
Unison.SF sounded very balanced between the instruments.
MagicSFVer2 sounded very nice (I like it, thanks to Dennis Deutschmann and whoever that made it happen.) but not balanced -- out of the 5 instruments, the violin sound was surprisingly loud. I end up changing its volume using the mixer. All other instruments use default (say 100 out of 127 full scale). But violin I have to reduce to say 20 out of 127.

1. I'd imagine the virtual sound volume would be balanced already. In other setups I've used, all the sounds tend to be balanced roughly. Is there some special things in terms of installation that need be done? All I did was edit preferences | IO and pick the SF file.
2. The developer of MagicSFVer2 suggest in its user note: start the creative "audio hq", click at the "Import Environment"-button and choose the "_Cathedral.ea2"-file. How can I do the equivalent of this in MuseScore? (Other than manually changing the reverb for all instruments to ~60%.)

These will be the primary SF I use, unless I hear suggestions from others. Unison uses a smaller memory footprint which I like, and MagicSF only about 30MB more with better sound, if I get the issues resolved... Thanks in advance for your help!


Comments

Balance between the instruments is the responsibility of the SoundFont developer. Unfortunately most of the free SoundFonts I've come across have these issues or are very large (which my computer doesn't have enough memory to handle).

My favorite SoundFont that I have been using for the past few months is hidden in Windows. On my computer it is: C:\Windows\Driver Cache\i386\driver.cab\2gmgsmt.sf2
You may need to copy the file to a new location since driver.cab is a compressed file.

In reply to by David Bolton

Well, we have 2 hypotheses here: one there may be a problem with MuseScore, another that the SF is the problem. So I thought: let's conduct some objective scientific experiments and see if we get somewhere...

I use a total of 7 SF2 files that were highly praised, and I found out that other than different quality of the sound (roughly proportional to the size of memory they take), their response to my score belongs to only two categories:
- all voices are balanced in sound: e.g. 2gmgsmt, Unison, and 2 others. They cover the broad range of smallest and largest memory usage, so no correlation there.
- Violin is significantly louder (by same amount); others 4 voices are balanced: e.g. MagicSFver2 and 2 others. It is interesting that they all perform the same way.

I hope this information may help us create another hypothesis and experiment further to get at solving the issue. Again, many thanks for taking time looking into this.

In reply to by newsong4life

Is there a way you could try your soundfonts in another score writer? That would confirm or eliminate MuseScore as the problem.
Perhaps the violin is loud in some soundfonts to make it soar above all the other sounds going on, a violin concerto for instance. In the orchestra this would be common usage for a solo violin, otherwise the violins patch would be used. But this would be inappropriate for chamber music. Just a guess on my part.

In reply to by MDMilford

I've performed more experiments, and am now agreeing with MDMilford and David Bolton. Thanks for your help!

Last time I got one result wrong. I've added one more SF, and all but 2 were showing the louder violins.

The 6 that were showing correct loudness were all reliable SF that had been in use for >4 years, most of them constantly revised based on user feedback (the ancient M$ version being an obvious exception). They all seem to record the violins at far-field (orchestra listening distance).

The 2 that were different were both based on near-field recording (closer than chamber distance), and so it was brighter and more strident, and louder too. I won't use those 2 SF for other reasons anyway, so I basically eliminated the problem.

With now >2 days spent on MuseScore, I'm more confident that MIDI import was causing most of my other issues reported elsewhere. (refer)

In reply to by newsong4life

Since you've tried so many soundfonts, could you post which one you found to be best for your purposes and a link to get it? is it free? I'm looking for another soundfont as I'm not very happy with the sounds of the Fluid sf I'm using. The cello takes too long to swell into it's tone so that staccato notes are inaudible, the bassoon is out of tune, and the oboe just sounds bad.

In reply to by MDMilford

There is a risk of sharing tentative results (as you saw in my last report that had to be corrected the next day), but here it is...

First, some comments. I am using my own test score, which is a contrapuntal free dodecaphonic composition of about 5 minutes, with 5 instruments (violin, flute, trombone, bassoon, piano) using maximum of 6 contrapuntal lines (simultaneously). I'm looking for SFs that gives me authentic sounds that can provide sufficient separation of instruments so dancers can hear all contrapuntal lines at the same time.

In general, the more memory a SF uses, the better sound quality it has, provided they spend enough time optimizing and improving on it year after year. Most of the SFs listed here has been in use for > 4 years, and improved upon regularly (except for the M$ version which was not changed since first introduced long ago).

More memory footprint typically loads longer, but not always. See last two. Also, there is a difference between the static (uncompressed size of SF) and dynamic memory footprints. Again the last two serves as best example.

You may think that the more memory the SF, they should sound closer and closer together. It's not what I've found. In fact, at the bigger memory range, it's sort of harder to decide which one to choose...

I tested with r1667 (0.94+) on Windows XP. Here are the ones that I found to have roughly nice balance of all 5 instruments, in order of progressively more dynamic memory footprint (actual memory used with MuseScore and after score has been loaded/played back), sound quality, and load time (except where noted):

Name Static memory footprint Dynamic memory footprint (with MuseScore)
2gmgsmt 2MB 58MB (this sounds exactly like Windows Media Player, for obvious reasons; and is also good for many other reasons...)
Unison 29MB 79MB
merlin_creative 21MB 82MB (compare static/dynamic with unison!)
merlin_audigy 37MB 99MB
merlin_orchestra 64MB 127MB
merlin_symphony 142MB 210MB (I like this best overall for generating WAV, only problem is it takes a LONG time to load!)
FluidR3_GM2 145MB 240MB (Notice this uses much more dynamic footprint, but loads very quickly! It's hard to compare whether this is better than merlin symphony or not. All depends on your perspective.)

When I use guitar as test subject, everything changes. I haven't had time to test each one, but the last two were pretty awful! I tested about 4 or 5 SF, and I actually like 2gmgsmt the best for guitar, considering its low memory footprint. I have a suspicion that I have to get a guitar-specific SF.

Sources:
2gmgsmt: Do a search on your windows PC, it's there since around 2001 I think. It's hidden inside a *.CAB in windows directory. You just need to get it out.
Unison and FluidR3_GM2: you can get it from MuseScore.
Merlin series: Note it changes nearly weekly, it has been around for a while, and there are far more versions than I listed -- most of them are highly rated. You can get it from composition-context.com
The good thing is: they are all free. At one time, there were some issues as to whether there may be intellectual property issues with FluidR3_GM2, but it appears resolved now.

Assuming there is no issues regarding copyright, it's easy to roll your own soundfont files using either 'Vienna Soundfont Studio' from Creative (http://support.creative.com/scripts/getprod.aspx?id=16559) or Kenneth Rundt's Viena (http://www.synthfont.com/). Both allow you to grab your favorite presets from various files, modified them in some ways (e.g., change the volume), and bundle them into a soundfont file. Voila! Your own personal soundfont! That's what I do.

In reply to by David Bolton

That Vienna Studio really doesn't need a Creative hardware in order to work - programmingly speaking, they just refuse to run unless you have a creative soundcard! There are lots of people like that in the world.

Kenneth Rundt's Viena (http://www.synthfont.com/) (note that it's not VieNNa) is an entirely different story. It works even if you don't have a soundcard at all on your system, and I really prefer it to the Creative program anyway. Give it a try David, I am sure you will be delighted. I have been using a couple of different personal soundfont files that I constructed through Vienna and Viena by picking and choosing favorite presets out of the nearly 4 GB of soundfonts I have collected throughout the years. My happiness (soundfont-wise :-) would be complete if I could find a good free male vocal preset!

In reply to by David Bolton

You are very welcome, David. I am surprised that Viena is not better known since most people can certainly make a good use of it. Rundt's SynthFont is also one of the best (and most versatile) Midi file players. It's what I used to create practice CDs for the choruses I sang in.

Maybe we should have a thread in the Forum to trade our favorite presets! :-)

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