I recently discovered the loop tuner in SoundForge ( I have SF 5.0)
Great tool, gives you much more precise control than Vienna SF editor.
However, I'm still running into some issues.

1) I'd like to make long loops since I have some fairly long samples
and want to take advantage of as much of the sample as possible.
However, it's difficult to find 2 points with nearly identical
characteristics at vastly different areas on the sample. It's my
understanding that the loop points need to be at zero points that are
part of areas that are essentially identical in appearance. Or is that
not correct? Hunting and searching through the sample in areas that
seem similar is tedious, is there a better way (maybe even automated?)
to do it?

2) Second, what do you do about looping stereo samples? I find that
it's virtually impossible on for example a string sample, to find 2
points where both L & R channels hit a zero point at exactly the same
time. What do you do about this?

3) Is it possible to have the looped section play for however long the
key is pressed, but when released, the sample plays out to its natural
decay?

Thanks for all input!

Doc,

Knud gave you some great advice (Knud, you da' man!), so there's only a
little I can add.

Hunting and searching through the sample in areas that seem similar is
tedious

Yes, it is tedious! It's even more of a pain with long samples like yours,
because there's so much to examine that spans a very large distance. If the
sample has vibrato it's a lot easier to loop, because you can just loop
between two or more peaks, or two or more valleys. Looping around multiple
vibrato cycles is usually better than looping around just one.

There's a program that claims to find the optimum loop points automatically
(called Zero Seamless Looper or similar), but the demo I tried a few years
ago was only slightly useful. So all you can do is search visually for
similar regions, and then use the CrossFade tool if needed as Knut said. Do
not underestimate the value of this tool! It really does a great job on
samples that could not otherwise be looped.

--Ethan

Yes, it is tedious! It's even more of a pain with long samples like yours,
because there's so much to examine that spans a very large distance.

This hassle is at least partially offset by the fact that notes rarely go
that long, and even if the loop isn't perfect, by the time it loops the
listener will have forgotten about the note since it has become part of the
texture. From a psychological standpoint, the attack portion is the most
important. Sometimes these big-whoopee high budget libraries waste gigabytes on
unlooped sustains while forgetting to provide useable attacks. I would rather
have 8 sustained string attacks with 6 second loops than 1 unlooped attack with
five hours of sustain!!!

This may be controversial but in my opinion sustain variance is sometimes
more important to realism than perfect loops. Say I have a 3 1/2 second flute
sample that can only be smoothly looped on a small portion. I'll often bite the
bullet and extend the loop to encompass most of the sample and live with the
timbral jump at the loop start. In an actual flute chord this sounds more
realistic than short loops even if it's ugly on a solo line! I've created solo
and chord version of the same samples, just by changing the loop points. This
way I can switch as needed. This only applies to "ghetto-style" sample editing
since ideally we would all be using long, pristine samples expertly performed!
In reality some of us have to eat.

Looping around multiple
vibrato cycles is usually better than looping around just one.

What's really cool is that you even if you can't find a spot that doesn't
click, a crossfade sounds almost like a vibrato hump, so you can just set the
loop points on either side of a series of humps and do a crossfade. Even pretty
extreme crossfade amounts are often imperceptible with vibrato samples.

Ethan, some day in that distant land called the future we need to record,
program and sell a bitchin' sample library.








blahblah

Knud,

I would rather have 8 sustained string attacks with 6 second loops than 1
unlooped attack with five hours of sustain!!!

Hear, hear!

sustain variance is sometimes more important to realism than perfect
loops.

Agreed. And we discussed a while ago the importance of being able to vary
the high-cut filter and other parameters while a note is sustaining to make
it more realistic. All anyone has to do is listen to live players to learn
what's important. Clarinet and other woodwind players constantly change the
timbre as notes sustain. With a real instrument the timbre and volume are
intimately related. As a note gets louder it also gets brighter. This is the
Number One reason samplers sound so phony for orchestra instruments on
legato passages. Even with a great multisampled patch, there's no way (other
than filtering) to change the timbre as a single sustaining note changes
volume. This is the main obstacle that must be overcome before samplers will
ever sound like real instruments.

Many instruments - especially brass - players start each note with a slight
accent, so the attack is clearly articulated but without the rest of the
note blaring as it sustains. This can be programmed in a sequencer, but it's
a tedious nuisance and the extra brightness is not part of the volume
accent. If you capture that attack as part of the sample itself the whole
thing is much more realistic. What happens in the first quarter second is
even more important for realism than what happens as the note sustains.

You know all this stuff and I'm not lecturing you smile. Just explaining
for interested lurkers.

some day in that distant land called the future we need to record, program
and sell a bitchin' sample library.

What I really want is to design a sampler that can modify the sound of a
Wave file in real time. It would do much more than just play back looped
samples. It would know how to cross-fade between different wave file notes
to better sound like one continuous line played by a solo flute or an entire
cello section. Versus sounding like a bunch of down bows all in a row. It
would use filtering to automatically tie volume to timbre in the correct
proportion. Yamaha's Vocaloid appears to do some of this stuff, but I think
it could be done much better than the examples I heard on their site. And
such a system wouldn't need 120 GB of sample data like the newer Giga
libraries. I'd think a few hundred MB would be more than enough raw data,
and probably a whole lot less than that.

Alas, if only I knew how to program DSP in Windows...

--Ethan

Knud,

Is it really so hard to have real-time interpolation between multiple
sources?

After hearing how well Sound Forge's crossfade looper works, I'm convinced
the same thing could be applied to multi-samples.

sample the waveforms at several hundred points along the way and
interpolate between them on the fly?

Yes, that's exactly what I've had in mind for several years now. And I bet a
lot of that could be handled with filtering, so only three or four separate
samples are actually needed.

--Ethan

Yes, that's exactly what I've had in mind for several years now. And I bet a
lot of that could be handled with filtering, so only three or four separate
samples are actually needed.

--Ethan

To state another possible approach: if enough dynamic layers were involved
(say a couple hundred at least) then a crossfade may cease to create
perceptible comb-filtering. However I think filtering would be a more realistic
solution, although I would say at least 6 layers should be sampled.


blahblah

lahblah (knud) wrote in message ...
multiple sources? With all the fancy DSP nowadays, emulating acoustic spaces,
analog processors... why can't someone program an algorithm to interpolate
between a 'bone played ppp and one played pp? Hell, why can't they take a
crescendo sample from an inaudible whisper to a blaring ffff, sample the
waveforms at several hundred points along the way and interpolate between them
on the fly? It has the potential to redefine what sampling!

hello all,

no, i think this problem is trivial and solved. essentially what you
are describing is known in the academic computer music world as
"waveterrain synthesis." there are implementations in cSound, maxMSP,
i think STK too...

spectrasonics "atmosphere" synth has an engine similar to what would
be required, but the core wavetables are not aimed at "realistic
instrument" synthesis.

i think, that the reason that this technique has not broken into a
commercial product, is that commercial electronic music products are
aimed more at electronic sounds, not recreations of acoustic
instruments. i'll guess that everyone who would care enough for
hundred-layer samples just hires real people to play instruments,
while everyone else just wants house beats and a sawtooth lead. brrrm
brrrm whhrrr whrrr. haha.

but in the mean time, you might try setting a random S&H or LFO to
modulate some parameters in order to introduce some pleasing
non-periodicity to the sound. for example, for a string ensemble
sample, a good choice would be to modulate the unison intensity. hope
this is useful.

Knud,

if enough dynamic layers were involved ...

Layers are great for different attack velocities. But to emulate an
expressive instrument like clarinet or violin you have to be able to start
loud and then get soft realistically, and vice versa.

--Ethan