Creating free sf2

[CheeseLord]

Over the last week, I've been thinking about my pitiful sample library. Although http://freesf2.com has been a godsend, the free department store is still lacking in several factors. Any solo non woodwind instruments are extremely synthetic, there are barely any articulations, rarely any velocity layers etc. etc.

But then, I remembered the Wesnoth Live Band. We have a multitude of instrumentalists in front of us, and we're letting them idle!! Why don't we create a bank of custom soundfonts WITH the articulations, WITH the quality, and WITH the velocity that libraries like EWQL possess. If the musicians agree, then is there anything stopping us?? I'm quite interested in seeing what others think, especially people that are cheap like me xD

Of course, I may be completely wrong, and we can never achieve commercial standards due to technological constraints with sf2, but even if it turns out to be the case, why don't we all give it a try

[West]

It's not a bad idea at all, though if we were going to do it there's no reason to aim for the sf2 format. Better to create the instruments in sfz format, which is more powerful, and then convert the sfz files to sf2 if necessary.

However, the biggest problems with doing something like this is 1) recording equipment and 2) the skill to use it properly. No matter how good someone is at playing an instrument, it's far from certain that s/he has the necessary gear to record the instrument in high quality. And even if s/he has, micing an acoustic instrument is an inexact science to the point of being a dark art, and if you don't have a good deal of experience with it, chances are the result will be mediocre at best. I have done a fair bit of micing things over the years but my recordings are still hit and miss. Most often miss, actually

And a good mic and preamp and knowing how to use them is not all: you also need suitable place to record. Either a treated room with barely no reflections or an untreated room that sounds good. Most of us have neither and trying to record something in a room full of weird peaks and nulls will add a whole other layer of difficulties.

So while I think this is a good idea, and I absolutely encourage everyone who feels like it to give it a try, be aware that sampling acoustic instruments is hard. At least if you want good results. You can't just point a cheap computer mic at an instrument and plug it into a generic onboard soundcard with $0.5 worth of components and expect it to sound good. It's just not going to happen.

[PPH]

I agree with West. Still, I think those interested might want to try. The results will be far from EWQL, but it's better to have samples of a real instrument that haven't been recorded in the best conditions than to have no samples at all or than having low quality, synthetic samples.

Also, most solo instruments have been recorded by the Philarmonia orchestra. Check their website. It might be a better use of your time to tweak these samples and trying to make sfz files out of them. This is tough work, but if you record instruments, you'll have to do it anyway.

[West]

I agree with West. Still, I think those interested might want to try. The results will be far from EWQL, but it's better to have samples of a real instrument that haven't been recorded in the best conditions than to have no samples at all or than having low quality, synthetic samples.

Yes by all means, do give it a try if you want to. I was merely pointing out that anyone who does try it shouldn't expect wonders unless they have good gear and know how to use it. Some quick tips:

* Record the instrument in minor thirds (e.g. on a cello, C, D#, F#, A) for as many octaves as it can play. It's less work than recording it chromatically, and loaded into a good sampler you won't really be able to tell that some samples are stretched.

* If your gear will let you, record at 44.1kHz, 24 bit. Otherwise record in 16 bit and...

* ...watch your levels. 16 bits have much higher noise floor than 24 and if you record the instrument at low levels you will get a ton of noise. This is not a problem with 24 bit recording.

* Don't record one note, stop, save the file, and record another. Just record all notes in one take.

* Record several takes, then pick the best sounding notes.

* As for velocity layers, three is probably fine for a melodic instrument (depending on what instrument it is, of course). Any more than that and you will just despair when the time comes to split all notes up and create an instrument. If this is your first try, you don't want to get in over your head. In fact, two will probably do as long as you make sure they're reasonably close in terms of dynamics. Having a sampled instrument that jumps from ppp to fff is very hard to use.

* Don't worry overly much about playing exactly in tune, as individual notes can be adjusted from within the sample format. Do however make sure to keep sustained notes on pitch, so that they don't start going flat or sharp after a little while.

[marcusvinicius]

Hello folks!
I think if you just want good samples you can find some here:
http://www.philharmonia.co.uk/thesounde ... s/library/
But samples are not just hard to record, they are also hard to mix, so you need a lot of experience to make good sf2 patches with those samples.
An interesting interview with a "sound artist", I think:
http://www.philharmonia.co.uk/thesounde ... n_chapman/
I hope I helped you a little,good job!

[PPH]

Hello folks!
I think if you just want good samples you can find some here:
http://www.philharmonia.co.uk/thesounde ... s/library/
But samples are not just hard to record, they are also hard to mix, so you need a lot of experience to make good sf2 patches with those samples.
An interesting interview with a "sound artist", I think:
http://www.philharmonia.co.uk/thesounde ... n_chapman/
I hope I helped you a little,good job!

These are the samples to which I was referring in my post

[marcusvinicius]

These are the samples to which I was referring in my post

I'm sorry man, I didn't realy see your post before, I was in a Hurry just answered CheeseLord's message.
It's good to know that people know that good resource that is Philarmonia's website.
good night!

[PPH]

These are the samples to which I was referring in my post

I'm sorry man, I didn't realy see your post before, I was in a Hurry just answered CheeseLord's message.
It's good to know that people know that good resource that is Philarmonia's website.
good night!

Don't worry. I wasn't complaining. Like you, I was glad someone else knew this resource. Besides, you provided a link, which I hadn't.

[CheeseLord]

Sry for not replying sooner- exams :s

West: I'm not sure what sort of recording equipment you need. I have a microphone, but that's all. Is there any chance of you passing judgement if I record a short extract??

PPH: Hehe, I already knew about that website. However, they've certainly increased their collection. The only problem now is the various articulations. Thanks!!

marcusvinicus: Same website as PPH posted :p The interview was very interesting though (although IMHO not entirely relevant :p , who cares!!)

My humble thanks for the replies and advice I've recieved. I'm going to begin recording fairly soon, and if this is a success, I'll start asking around. Maybe one day, I'll actually get a professional quality soundfont

[West]

West: I'm not sure what sort of recording equipment you need. I have a microphone, but that's all. Is there any chance of you passing judgement if I record a short extract??

What kind of mic is it? I.e. brand, model, type etc.

[CheeseLord]

Hehe...i thought you might ask that

It's a ATR25 and here's a random spirited away piano thing to test quality: http://www.box.net/shared/91pt08fng3 My bro got it originally for RAM, but oh well...

I eagerly await your judgement

[West]

Well, it's an Audio Technica which is a quality brand AFAIK. I can't really tell if it sounds good or not when the recording is 22kHz, 16 bit and peaks at -29dB (not counting the loud click in the beginning). This is basically an example of how you SHOULDN'T record something with a mic



Like I've said, if you need to record at 16 bit (have you made sure you can't do 24?) you must make sure to record at good levels, because 16 bits has a high noise floor and below a certain dB level your recording will be a wasteland of quantization noise. I'm not kidding, this is important. If you're going to record at -29dB, you're actually doing just an 8 bit recording or something, because that's all the resolution you're getting with such a weak signal.

Secondly, make sure to record at 44kHz. This is cd quality and it makes no sense to record at a lower sampling rate.

[PPH]

West is right. The bit depth (8 bits, 16 bits, 24 bits) determines the maximum precision of the recording. Digital recordings are sequences of numbers that represent the height of a sound wave at fixed intervals. Each one of these numbers is called a "sample" (because you are not recording the whole wave, but just samples of it, that is, parts of it at fixed intervals). As waves are continuous, they can only be represented with real numbers. So, when you make a digital recording, these numbers must be rounded. The greater the bit depth, the more values you can represent and therefore the smaller the rounding error. Rounding errors are random. So, high rounding errors are like adding random noise to your sound. Therefore, the higher the bit depth, the less noise.

The fixed intervals at which you record are determined by the sampling rate. If the sampline rate is 22Khz, you take a sample each 1/22000 seconds. The sampling rate must be double of the frequency of the wave for it to be possible to reconstruct the wave from samples. That's why you should record at 44 KHz: the human ear can hear frequencies of between 20 Hz and 20KHz.

[West]

Thanks PPH. Great explanation! And to anyone reading I might add that the rounding errors PPH are talking about is the same thing as the "quantization noise" I mentioned above. This is generally considered something undesirable unless you're into oldschool lo-fi sampling á la C64 music.

CheeseLord, what me and PPH are talking about here are digital audio recording fundamentals. This is stuff you should know before even thinking about clicking that REC button. Sure, trial and error can be educational and sometimes even fun, but if you want good results and don't like wasting your time, I really recommend familiarizing yourself with these concepts.

[Dratz-_C]

Hello,
You probably know this information, but http//:theremin.music.uiowa.edu/ has some chromatic recordings of instruments in 3 velocity layers, some in vibrato and non vibrato versions. These samples are free for any use. I was the one that created the UoITrumpetNV4 SoundFont from the trumpet samples available there. I did this manually in Audacity, AutoTune, and Viena -- http://audacity.sourceforge.net/ , http://www.analogx.com/contents/downloa ... eeware.htm , and http://www.synthfont.com , respectively. Sometimes the UIowa site is down, as it appears to be now; however I have contacted the administrator and predict that it should go up soon.
Also, I am working with the creator of SynthFont and Viena to add some new features such as velocity blending, so a pp and ff set of samples can be cross-faded between each other. If any of this progress interests you, look at and feel free to post about it in the SynthFont Forum, or here. http://ts-central.com/tsbb/?CategoryID=2
Cheers,
Chris Dratz