What I don't know about soundfonts

[Leonardus]

Not too many years ago-it was pretty much accepted fact-soundfonts were destined for extinction. Soundfonts were limited in some way (8-bit?) that emergent Sampler formats weren't.

Somehow Soundfonts survived.

What happened? Can I get an excellent sampled Piano and Drumkit in soundfont format, something rivaling the industry best?

How does the future of Soundfonts look?

Is Gigastudio king? thx. Leonard

[birrbits]

well, soundfonts certainly aren't dead. Nor are there any massive limitiations such as you described that would prevent many outstanding soundfonts from existing today. There are some truly excellent piano soundfonts out there, some free some pay.

[loomchild]

In many regards, soundfonts are very comparable to the mp3 format. They are using a pretty old and somewhat outdated format which has been bested by most other formats in the last few years but which is nonetheless widely compatible. Any soft sampler provides soundfont import.

[Hovmod]

Try this: http://www.naturalstudio.co.uk/ns_kit7free.html

[SampleSmith]

Leonardus, check this out:(The sfz thread)

http://www.kvraudio.com/forum/viewtopic ... sc&start=0

This one of the best reads I've had about soundfonts. 96 pages over 2 years on KVR (I'm only in about 8 pages at time of writing). I wish I'd found it before today.

Seems they are far from dead, and there's an embarassing amount there for me to learn. I hope it's helpful to you too.

Keith

[Snoogins47]

I don't know for sure on all these counts, but I'm fairly certain that Soundfonts are limited to 16-bit, 44.1k, and Mono. The mono thing is fairly easily worked around (I think Vienna even splits stereo files up into two mono files when you import them) and the other two are, for most people, far from crippling. Loom's comparison to mp3 is fairly good: soundfonts aren't at the leading edge of sample formats, but they're of a high enough quality to be very useful, while being almost unquestionably the format with the widest support.

I would hesitate to say that you'll be able to find soundfonts that "rival the industry best," with how extensive the "industry bests" are these days. You won't find a soundfont that has anywhere near the detail of these massive multi-dvd, 4-figure pricetag jigs these days, but that's not inherently a bad thing. A lot of it depends on what sounds you're looking for, and what you want to do with them. A drum soundfont probably isn't going to rival BFD for realism in the acoustic drumkit realm. Squidfont orchestral isn't going to trick your audience into thinking they're listening to the Chicago Symphony Orchestra on the same level as say, EWQLSO. But with the myriad soundfonts available, and the fairly versatile nature of the format, I would say it's still very far from extinct.

[learjeff]

You're right about the bit depth limitation: 16 bits. However, there's a field to encode the sample rate, and good players should honor it by compensating, or at least warn if the sf2 file sample rate doesn't match the current operating sample rate.

sf2 is also not limited to mono. For example, my free jRhodes3 soundfonts includes a stereo soundfont and that's the version I use for live playing.

The big advantage to the soundfont format is its ubiquity; it's the best format to find free stuff by far. The biggest limitation IMHO is that you can't use velocity to directly control filter parameters; instead you have to program velocity zones. This creates a jump at the zone boundaries when contiguous zones use the same sample and the boundary exists only to modulate filter parameters. This is more of a hassle for the sound designer, but due to the technical nuisance it is, many amateur soundfonts that might otherwise sound good don't, and fixing it by patch programming is tedious.

But there are a number of other limitations, such as inadequate support for release samples (they're there, but not controlled properly based on limitations in the sf2 spec) and lack of a convolution engine (for modeling a piano soundboard, for example).

I agree that sf2-format offerings don't rival big bux sounds, but the reason is more a matter of marketing than technical aspects. Not entirely by any means, but mostly.

A lot of fuss is made about the 16-bit limitation, but I doubt it's significant.

[93143]

Soundfont 2.1 has a fairly extensive mod matrix, including velocity-to-filter, and Soundfont 2.4 can handle 24-bit samples. Unfortunately, the spec everyone supports is 2.0...

I've seen release samples done by looping a small section (64 samples maybe) of silence at the beginning of the sample and turning off release looping. I use this method myself now, because it allows the release to be affected by the envelope/LFO state at note off.

[janp]

I like to tell you that there's a new soundfont spec :sf2.4
wich is 24 bit. The latest version of vienna is up to this.And if you don't have a SB cart you can use the latest version of Awave for this.
It's not dead IMHO.

Jan

[SampleSmith]

Just to clarify: Vienna is at 2.4 The current soundfont spec(according that version of Vienna) is 2.1 (I believe it was 2.01 for ages, which adds to the confusion.)

I guess the only thing to do is try and make a 24 bit font and run the results through a bit meter?

[Leonardus]

Thank you, everyone! I appreciate all your input.

I have a question.

Of dedicated Soundfont "Players [?]", are there significant differences among them? What factors should I consider when choosing a dedicated Soundfont Player? Do Soundfonts sound different player to player? Is CPU load an issue among dedicated Soundfont Players? Thanks. Leonard

[Snoogins47]

sf2 is also not limited to mono. For example, my free jRhodes3 soundfonts includes a stereo soundfont and that's the version I use for live playing.

I already can see that you're much more well informed than I am in this respect (I'm a huge fan of the old EP sound, so I discovered your excellent soundfont some time ago, and it's probably in my top 5-10 most-used fonts) But, the stereo/mono thing was something that I was almost positive on, and so I wanted to dig into this a little more deeply: whenever I've attempted to put together a soundfont from stereo audio files, it gets split into two mono files. And in fact even when I load up your Rhodes soundfont in Viena it shows the same thing: two files for each sample, one marked with L, one marked with R. The end result is a "stereo soundfont," but it's formed using mono samples.

I will also mention that this could be due to something on my end: some setting, or some weirdness with soundfont import, or who knows what else, that may cause this on my box and not on others... but since I seem to recall it happening on at least two boxes with different cards over the years, I don't know.

This is what I was trying to communicate in my first post, which I probably failed to do successfully, so, my bad. The difference is probably somewhere between "nitpicking" and "effectively the same" if we're talking about the end result, but...

I haven't done extensive testing with the format (as soon as I started to have some troubles, I ran directly to SFZ ;P) it seemed to me that this, while a very subtle difference in many respects, I seem to remember (though I could very well be wrong) having some severe voice stealing issues with a stereo drum kit I tried using on an older machine. I didn't look into it a whole lot, but after I spent a bit of time unable to figure out what was causing the problems, I decided it was quite likely that I ran up against the polyphony ceiling twice as fast as I would have in a 'mono' font. I never really sought hard, fast answers on the issue, since I was already well on my way to figuring out yet another way to work around something (story of my audio life!)


Leonard: There are others that can answer this question better than I can, for sure. I currently use SFZ almost exclusively for Soundfonts (though I like to mangle them in Crystal sometimes ;P) despite having a SB card that directly supports Soundfonts. My efforts to use soundfonts through the Audigy have been about as entertaining as getting all my teeth ripped out simultaneously with dental floss, but not quite as productive.

I would imagine that soundfont players, as a rule, aren't going to sound too drastically different from one another on extensively sampled instruments, since it's pretty much just simple audio playback. Instruments that require resampling, however, may make this a significantly different story, and since I have very limited experience with almost everything other than hardware, and SFZ, I can't really guide you too much in this respect.

Despite the fact that this post, like most of mine, has turned into a novel, one thing I wanted to mention was reminded by something Learjeff had said.

If you're willing to put in the effort, there are often soundfonts you can find on the 'net, where the samples themselves are nice, but the end result is mediocre at best, sometimes completely unusable. If you're willing to put in the effort to fix all sorts of problems in the mapping/programming, you can often find some 'diamond in the rough' soundfonts. Problem is, you can often end up spending hours fixing a soundfont so that it goes from sounding like shit, to sounding like crap. Not really worth the effort.

[Leonardus]

Thanks, Snoogins47.
I downloaded the Soundfonts from the links provided and read through about 7 pages of the SFz thread before skipping to the end- page 97. Quite informative.


Which Player works best? What should I look for in a Soundfont Player?

[learjeff]

Start out using sfz, which is simple, free, and great. Where you go from there depends on your needs.

Snoogins, there's no difference between two mono samples panned apart and a stereo sample. In other words, Viena shows them as two mono samples, but other editors show them as a stereo sample. It's all just a matter of interpretation. You can avoid the issue by using "sfz" which supports stereo audio files.

Yes, you would run into the voice stealing issue twice as fast with stereo. This is true regardless of the internal representation.

[learjeff]

Someone please guide me to a copy of the mythical Soundfont 2.1 spec. When I follow the link labeled "Soundfont 2.1" at http://www.soundblaster.com/soundfont/faqs, it serves up the SF 2.01 spec, of which I have a well-thumbed copy, and which does NOT support 24-bit samples.

I see that for SB X-fi, it says "You will also enjoy unprecedented hardware synthesis capabilities with 24-bit SoundFont sampling!"

I suspect this is Marketing trying to make something more than it is. I bet that the output of the sounfont player is 24 bits, which is a good thing. Anyway, yes, try to use Vienna to build a soundfont with 24-bit files and see what happens!

I stand corrected about velocity. Somehow I got convinced earlier that SF2.01 didn't support velocity controlling filter cutoff, but it's right there in black and white, I've very happy to find. Thanks! It's a very important feature for sample sets with few velocity zones, especially for instruments like piano and guitar. Rhodes doesn't need it or I'd probably have been better informed.