Home Grown Default SoundFont for MuseScore

• May 6, 2012 - 14:02

There have recently been discussions about the shortcomings of the TimGM6MB soundfont in the issue tracker, and I have been looking around for a replacement.

After downloading and testing several candidates, I have come to the conclusion that the only real alternative to using Tim BrechBill's soundfont is to produce one collectively from the MuseScore community, because there are problems with all the potential candidates to replace it - either due to sample copyright issues, or faults in playback such as tuning.

The way it would work is this: We would use samples recorded for the purpose by the MuseScore community, and with either a Public Domain or Creative Commons licence. These would then be collected by the SoundFont development team, processed and then incorporated into a soundfont specially made to be compatible with MuseScore, and its particular needs.

I have sample manipulation skills (and tools) from a prior existence as a MIDI programmer, and would be prepared to co-ordinate the SoundFont project, should the community decide this is a good idea, but I will need the assistance of other people with similar skills.

I know that some members of the MuseScore community have already made steps towards a custom SoundFont - the Drumline percussion immediately springs to mind - and we could incorporate their work in the project.

If you would be interested in forming a SoundFont development group, initially reply here, and if there is enough interest, we can establish a home somewhere on the net.

Comments

• May 6, 2012 - 15:50

Awesome idea ! Some remarks.

TimGM6MB.sf2 is the default soundfont for MuseScore on Mac and Windows while for example on Ubuntu, MuseScore is linked to a package containing FluidR3_GM.sf2. Fluid is also 100% free, it's a lot better than TimGM6MB.sf2 but it's very big. That's why we used TimGM6MB.sf2. Even the modified soundfont with Drumline percussion is around 30Mb, 5 times the size of the default one. So, the first job of this "task force" would be to define a sensible size for a default soundfont or produce several ones with different size and quality.

Second, there are already a couple of projects to collect free samples. These samples could be reused.

  • Freepats: Freepats is a project to create a free and open set of instrument patches, in any format, that can be used with softsynths
  • Freesound: Freesound aims to create a huge collaborative database of audio snippets, samples, recordings, bleeps, ... released under Creative Commons licenses that allow their reuse.
  • Sonatina Symphonic Orchestra: a creative commons-licensed orchestral sample library

Reply

• May 6, 2012 - 16:05

I recognize the need for a small package for people who have limited internet speed/access. OTOH, most people are running on machines with 500GB+ of storage and internet of reasonable speed. Look at what is being downloaded- large video files in the GB region, not to mention streaming video, even large programs, music albums which in their compressed format can still total 100MB.

Why would compromising a sample library constrained by size be seen as a good thing?

Reply

• Nov 20, 2012 - 20:47

Small means faster on any given hardware.
In a previous life I worked for a well-known hard drive manufacturer that produced bleeding-edge hard drives in terms of capacity and performance. A running joke among the engineers was to send a few samples to Microsoft in Redmond, Washington to see if they could fill it up. Invariably, code bloat would produce exactly that result. I've seen late-2000s software perform as slow or slower on 2-3 GHz machines than mid-1980s software on a 4.77 MHz IBM PC due to code bloat. I've also seen a MIDI player use 70% of my computer's 3 GHz CPU capacity just playing a tune through a software synthesizer, leaving the keyboard and mouse jerky and unresponsive. Size matters, even in 2012.

Reply

• Nov 20, 2012 - 21:37

Granted. But the point is, there is a tradeoff between size and quality. And we already have a plethora of low- or average- quality soundfonts to choose from if you are trying to optimize on sound. What we lack is a very high quality soundfont to use in the cases where quality matters more than size.

Reply

• May 6, 2012 - 16:50

First, thanks for getting the ball rolling, and count me in any way I can help.

My peiorities would be a bit different, but not necessarily incompatible.

The current TimGM6mb soundfont isn't bad for what it is - a GM soundfont that can fit in 6MB of space. However, being limited to just GM sounds is an annoyance, as there are any number of instruments that people might want to write for that are not represented. And any number of insteumental effects - like slurring for winds, or more mute options for brass - that as far as I know would need sample support to implement. Plus, of course, as mentioned, many if not most would gladly pay a price in disk space for more realism in the samples.

So most important thing to me wouldn't be a new low end soundfont, but a new high end one. Howeever, it seems plausible that once the basic samples are assembled, the soundfont could be made available in two versions, so the soundfonts would be similar in all important ways - the patch assignments for the non-GM sounds, the relative levels between instruments, etc. I would guess things like just including fewer sample points and velocity layere per instrument, mono versus stereo samples, maybe also reduced bit depth or frequency for the smaller soundfont - anything that would allow a smaller soundfont to be built from the same basic source samples.

But if the effort is just to produce another relatively low quality 6MB soundfont that anyone who cares about sound would immediately replace with FluidR3 or the like, I can't see it being worth the effort. I would rather we concentrate our effort on the high end, and then let a new low end soundfont fall out of that but in a compatible way (so instrument and mixer settings developed with one would work with the other).

Reply

• May 7, 2012 - 14:34

I think the important thing is the production of soundfonts compatible with Musescore's needs, thus any quirks of the customised Fluid synth can be catered for.

So, whilst the production of a small size soundfont for inclusion with the Musescore download, probably consisting of mono 8 bit 22KHz samples for space could be the initial project, there is no reason at all why other, better quality soundfonts could not be produced, particularly as 2.0 allows the use of several soundfonts in a piece.

Perosnally I would like to see a Brass set with several velocity layers representing different attack capabilities of those instruments, and an organ set which is made from real organ pipe samples, not the mix and match approach of Jeux.

Reply

• May 16, 2012 - 01:14

I'm no expert on soundfonts, but was thinking about the wide variety of samples that must be evaluated to find the "best" ones. Am thinking that there must be a set of objective/technical criteria (e.g. sample rate, layers, tuning, no-clipping) as well as subjective criteria (e.g. how does it sound?) for each set of samples. If the list of objective criteria is agreed upon, then a set of tools could be developed that measure each set of candidate samples to see how they stack up. Any samples that have objective/technical problems could be flagged before being subjectively evaluated (by people). Does it make sense to list out the the objective criteria?

regards,
markt

Reply

• May 6, 2012 - 17:26

Oh something I missed in my previous comment. I don't think it's a good idea to host the work and discussion about this effort on yet another website. MuseScore.org could probably give some structure and it has the logging system in place.

Reply

• May 6, 2012 - 18:12

I once brought this up too (link ).

I think it would be good, because everyone can have their say in what they need from a soundfont - it would perhaps be better maintained too.

Reply

• May 10, 2012 - 08:57

On my Windows machine I've had excellent results when converting MIDI to MP3 or WAV format, using a free program called KarPBO, by Arthur Agapoff. The URL is: http://artagap.free.fr/karpbo/index.htm. The page is in French, but it's fairly intuitive what you have to do.
There are two soundfonts listed on that page: Chorium (27.5 MB) and Titanic (downloadable in 5 archived parts: 4 × 30 MB + 1 × 27.3 MB). I'm going to give them a try with MuseScore running on Kubuntu 12.04.

Reply

• May 15, 2012 - 12:40

I tried this sound font - Chris Crisis... I found serious tuning issues. and a general lack of sustain in brass instrumentation.

I subscribe to the view that the standard font that ships with MuseScore is more than adequate as a base.

What may be useful on first start up of a freshly installed MuseScore system would be a friendly system check to assess if the machine is capable of easily running the larger MuseScore font (whatever that may end up bieng) ?

Additionally, whatever font is used I would welcome more control in the mixer of the individual voices.
For example, to create a Cornet sound I currently combine duplicate staves of Oboe and Trumpet... both voices together more or less emulate a Cornet (due to the brassyness of the Trumpet and vibrato of the Oboe).
If I had more control of the tonal shaping of each instrument from the mixer that would be Great !

Thinking along the lines of the above:-
What about the ability to assign 2 voices to the same stave ? ... Possibly the combination of standard GM voices may address many peoples needs ?

Lastly.... and this may be way off on a tangent, I would really welcome the ability to generate sub voicing as a sub mixer section below a Voice.
For example..... A brass band has a total of 10 Cornets.
4 of these known as Solo Cornets who all play the same part.
What I would like to do is write one Solo Cornet part then be able to duplicate the voicing 4X at the mixer ... So under the Solo Cornet mixer entry I would see 3 sub mixer voices.
Maybe also the ability to stager the transport of each duplicated channel a millisecound +/- which helps to generate realism. (a pro studio production technique)

Much of what I propose is independant of any sound font developed, but could be a way of making more use of anything we do develop

Reply

• May 15, 2012 - 15:47

Not sure what you mean when you say the current soundfont is fine "as a base". If you mean, as a small soundfont to ship with MuseScore, I agree, but you can't build a higher quality soundfont out of lower quality samples. If we are going to build a higher quality (which is to say, larger) soundfont, there is no getting around the need to start from better samples.

Reply

• May 15, 2012 - 18:03

Maybe what we need to do is start on a specialist SoundFont for Brass Band, as the mechanisms for doing what you ask are not yet there in the MuseScore sound engine, but can be done within the SoundFont.

4 Cornet samples could be layered together to produce the sound you're talking about, and you could have the other brass band insturments available in solo and chorus versions.

Would you be able to assist in providing good quality samples?

If so, perhaps we should start a new thread to save this one becoming even more convoluted :)

Reply

• May 16, 2012 - 13:14

I would be happy to assist.
I am a multi brass instrumentalist.
I play in both Brass and Big Bands.
For example, one of the UK's leading Orchestral Trumpet Principals sits next to me so I should be able to get some good Trumpet samples at the very least...

Here is Me playing a Cornet http://sites.google.com/site/freemusicforkids/free-music-news/onbonches… ... A Bit wobbly with too much Vib... however, with samples you only have to get it right once.

I am also involved with live recording - Owning some respectable studio condensers and portable pro recording boxes/compressors/mic preamps etc.
Here is the Royal Scottish National Orchestra Brass Quintet recorded live. I used a overhead Jecklin Disk on a rig within a very ringy town hall. http://soundcloud.com/its_jon/rnso the latter half is more exciting :)

Whilst I can't boast an Anechoic Chamber ... I believe I can contribute some very clean samples to the project recorded in suitable and very dry locations.

I assume it is the absolutly most generic instrument samples of all catagory's we aim to capture.

I will need some guidence on the format/legnths these samples should be created in.

Im currently using Ardour and Audacity ...
Have been involved with the Deliplayer project when it was ported to PC .. testing and UI/GFX etc. http://sites.google.com/site/freemusicforkids/-continued/deli100%5B1%5D…
Oh... plus I ran a very high end electroacoustic buisiness at one stage in my life dealing with Scan-Speak, ATC, Dynaudio components within my cabinet designs. http://sites.google.com/site/freemusicforkids/-continued/labarinth.jpg?…

Through the pages of FMFK http://sites.google.com/site/freemusicforkids/home I could also contact the wider Brass Band community to locate the best voicing I can find... If you will allow me to do so.

Quite exciting :)

Reply

• May 20, 2012 - 00:59

Im happy to provide original samples and/or aid to engineer/adapt brass samples gathered.
I do a lot of sound editing, so used to maximixing what is important from a sound.
Ensuring the 'sample set' is unified as a collection is of equal importantance to the realisim of individual voices.

For me, getting a generic set of brass sounds that cover the range of instruments in place and working tonally to the same standard as FluidR3 would be priority one... Then, to intergrate a second layer of brassy realisim.

From a personal point of view... 2 instruments always sound poor.. Flugal and Cornet. None of the typical voices found in GM appear to come close to either.

If you have any samples/fonts you want me to test or rework.... send them :)

Does the project have access to a FTP area or similar ?

Reply

• May 21, 2012 - 17:30

Well, as you are aware using even mono 44.1kHz uncompressed samples can run into several hundred MB per sample.

Ideally I would like to have a minimum of a terabyte at our disposal, however, we can look for that later. For the time being I would say 10GB is a reasonable ballpark figure.

My own webspace is limited to 2GB so that's a non-starter. I would like to set up a dedicated ftp server for the project, but I can't do that for a few months - I'm moving again, so will need about 3 months before the dust settles on that.

Starting a new thread on this subject.

Reply

• May 11, 2012 - 17:05

Since we now have an actual forum for this topic, I am going to take the liberty of starting a couple of new threads for discussion of specific aspects of this project, as otherwise this one is going to get hard to follow very quickly.

Reply

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