SF2 loop problem - desperate!

[zircon]

I've been working on a song for quite awhile now, and at the very end I introduce a pad from a soundfont that to me seems pretty unique. I like the sound of it and nothing else I have sounds like it. However, upon rendering the song I hear some clicking in the right ear - on further investigation, it seems like a certain range within the soundfont was not properly looped. The soundfont ("BIG Pads Demo Bank V1.1 Oct 98") is being loaded in FL6's Fruity Soundfont Player, for reference.

www.zirconstudios.com/coolpad.mp3

The notes between and including C5 to B5 are problematic. Everything above and below are fine. This, however, is extremely confusing as when I opened the sf2 in Kontakt 2, I only saw two WAVs being used in the patch, and the same WAV that was within the C5 to B5 range was also above it. Why would the loop click in that range but not in another? I actually double checked the loop point in Zero-X Seamless Looper and it's absolutely flawless.

So, I'm thinking "maybe it's an FL problem" and open it up in both K1 and K2, but neither program loads the patch properly. They screw up the loop points further, as well as the LPF built into the patch, making both programs totally useless for proper loading of the sound.

I'm right on the verge of finishing this song, and this one stupid pad is holding me back. Can anyone help me fix this problem? I'm really desperate here.

[experimental.crow]

tried sfz ?...

http://www.kvraudio.com/get/769.html

[Spe3D]

Unlike velocity layers where you can safely get away with one - when it comes to pitched instruments that span a keyboard range - its always best to have as many samples as possible (at least one for each note played) - or with restricted playback to the sample pitch range maybe one or two notes either side its native pitch.

What happens with sound fonts with few samples that go a long way outside of its range is, as the one sample that now has to cover many notes, the higher up the keyboard you go, the faster the playback becomes (this effectively makes the sample shorter) - the lower down the keyboard you go, the slower the sample playback (this makes the sample longer) - now this is not too much of a problem with samples that have really big loop points, but it is with samples that have short loop points (could well be perfect at its native pitch, but sure will show up if covering notes it cannot handle)

I can probably make this clearer, but hopefully you get the idea.

In short its most likely the sample has a very short loop and being made to play outside of its native pitch.

Best regards,

Spe3d

:O)

[pettinhouse]

You can use a Vinyl scratch noise to cover the click!

[pettinhouse]

If you have Kontakt, import the soundfont in it and Edit with the Kontakt sample editor the loop using the crossfade function. If you don't have Kontakt you can import the samples in Soundforge (WIN) or soundtrack (Mac) and edit the loop there

[matthewjumps]

or - load up the sample in FL Studio as a standard sampler - use the elastique tonal pitch shifter to shift it up - render of a new wav, and do it a few times till you have a bunch of new wavs with higher pitches - then use a soundfont editor like kenneth rundt's viena to add in the new samples....then you can loop em properly and you will have less problems cause you will less resample pitching to do....if u know what i mean

[Sascha Franck]

I second the crossfade option in Kontakt. This would be the quickest way to adjust things. Do you have a link for the patch so I could give it a try?

[bro. miles]

Since you mention having Zero-X Seamless Looper, you could also use the crossfade loop function there. If the loop is very short, another thing to try would be making the loop 2X or 4x as long.

[Klemperer]

The soundfont itself, as it seems (had it once myself) should be the one HERE. Not sure if it's version 1.1 though. If anyone wants to have a look at this and doesn't have SFARK (needed to extract the soundfont) - the link for that follows directly to the sf2 itself on the page.

[zircon]

Thanks for all the input; I spent several hours last night trying everything I could think of, but ultimately, the last thing I did was load it into DirectWave. Believe it or not, it worked there. The LPF and filter envelope didn't translate perfectly, but there were absolutely NO problems with the loop points. Given that the two component WAVs of the soundfont were looped flawlessly to begin with, this leads me to believe that the problem was exclusively with the FL Soundfont Player and nothing else. Of course, Kontakt (both 1 and 2) could not load the sf2 properly either, so perhaps it's just an ornery patch.

[BertKoor]

the problem was exclusively with the FL Soundfont Player and nothing else.

You can try an alternative SoundFont player, like SFZ.
http://www.kvraudio.com/get/769.html

It's one of the best around, and for free....

[learjeff]

I've noticed that some soundfonts have the flaw that they don't retain 20 words of zero following the sample data, as required by the sf2 spec. This can cause clicks or not depending on the amount and direction of pitch shift.

Fixing it is fairly easy, if you're a nerd with a binary editor and the SF2 file format on hand (if there are few samples, that is). I'd have to think about the best easy way to fix it.

If just running it in sfz doesn't fix it (and sfz is subject to this problem), you could try any of the conversion programs and convert it to 'sfz' format and the problem might go away.

A free demo conversion tool to try is CDExtract (google it).

[bro. miles]

learjeff you just nailed the problem. Several ways of fixing this come to mind. Basically pasting either silence or another copy of the looped section after the loop, in any wave editor, would do it. It doesn't have to be zeroes, but it won't loop right if the loop end = sample end. I know for sure it doesn't have to be zeroes because the sf2 spec allows release data after the loop as an option. BTW your EP rocks (or jazzes - is that a word?).

[Samplecraze]

Bias Peak used to require 128 cycles before it could even locate the start point.

Emu Eos based samples only needed 5 cycles.