Hand Held Sound Flying Hand Percussion - awesome

[stubbsonic]

I think you have both misunderstood my point. You would not be getting rid of the dynamic range, you would be both increasing it and smoothing it out. I'll try to explain it more clearly.

Option 1 (HHS's approach): NO NORMALIZING, NO VELOCITY SCALING
Here you have some number of layers, each layer represents one dynamic level. Let's say you have 10 layers; quiet samples at layer 1, and progressing up to loud samples at layer 10. That means you have 10 distinct steps of loudness levels. If you did fast dynamic rolls with cresc/decresc in them, you'd hear it step from layer to layer. I suppose round robin samples if the hits are not precisely similar could mask this stepping effect, but it might cause it to sound less controlled.

Option 2: NO NORMALIZING, SOME VELOCITY SCALING OF AMPLITUDE
This is the thing I was expecting at a minimum. You would still have the 10 layers of velocity switching (in our example), but each of those layers is ramped/scaled (within it's particular range) by some degree of velocity controlling the amplitude. But as I was saying this would be messy to program. It is avoided altogether in Option 1.

Option 3: ALL SAMPLES, NORMALIZED, ALL VELOCITY-SCALED AMPLITUDE
Here you still have the same 10 velocity-switched layers, but with ALL samples normalized to full gain. NOW you apply velocity to control the dynamics of all the layers. A note with a velocity value of 2 will play the sample in Layer 1 very very quietly, a note with a velocity of 8 still plays Layer 1 quietly, but less quietly. A note of velocity 14 has moved to Layer 2, but now louder than velocity of 8, and will ramp upward until the velocity moves to the 3rd layer and so on. Now, rather than 10 steps of loudness level, you have a smooth 127 steps of level as you increase the velocities of the notes you play. You still move between the velocity layers but in ramps that you can control by adjusting the velocity curve and mod depth. In other words, the samples of the quietest hits are first normalized, but then made quietest again by the velocity/amplitude scaling. The advantage is that with velocity controlling the dynamics you get a perfectly smooth (or at least very good) dynamic transitions between the samples. From very soft to very loud.

Option 4: RIDING/BOOSTING LEVELS DURING RECORDING, NORMALIZED, VELOCITY SCALED AMPLITUDE
With this approach, you are simply trying to get some of your clean gain while recording, so you are using a nice mic pre for clean gain while tracking. The advantage with this approach is that you avoid having to normalize a very low level recording (which might be only using 13-bits or less) up to 24 bits. In that case, you would not get back any of that resolution through normalizing. Riding levels while tracking lets you keep better resolution on the quiet samples. But as I said, since those quiet samples will ultimately be played back at very quiet levels, it doesn't matter as much-- except when the drums are processed with lots of aggressive compression (which pulls up the quiet samples and makes that stuff a little more audible).

The advantages to the last approach are proven in the stuff I've made. But it does require very careful performances, slating, and editing to get it all to work right. The other way is simpler, but pretty limited.

Hope that wasn't over-explaining. I just wanted to make sure my point was clear. Seems like some sample library developers don't get this.

[Shooshie]

In the last option, riding levels, do you hear any trouble with background noise as you crescendo? I guess it's as you said, low enough playback levels that the background noise is compressed out. I just keep imagining it, and in my imagination I hear background changes. But as long as those quiet levels (recorded at higher gain) are played back quietly, and it shifts to layer 2 before reaching a level where the background becomes audible, then layer 3, and so forth, then it probably is never audible. I'm just over-thinking it, that's all.

Shooshie

[monkey man]

I thought that the angle was that instead of introducing artificial dynamics (mapping normalised files to velocity), the samples are instead mapped directly to velocity, thus preserving the natural progression as played by the drummer.

The only difference I can see, apart from possible artefacts introduced by the normalisation process, is that the artificial mapping (formerly-mentioned method) is a process borne out of theoretical perfection, that is, an even transition envisaged by the programmer, whereas the latter directly links the dynamics played to velocity, human variations intact.

In other words, the volumes of the relative dynamic samples may not increase in a perfectly-linear fashion in real life, whereas when normalised, mapped to velocity and controlled by the VI / synth's amp engine, said levels would surely follow a more smooth path from quiet to loud.

EDIT: Didn't realise there was a 2nd page.

Stubbsonic explained the options concisely. Thanks mate.

The thing is, that, using your example, two different strike-levels' being represented by the same sample, albeit with different amp levels, isn't realistic. In reality you'd hear differing timbres between those two strikes. I'm guessing many more than 10 layers have been employed, ensuring abrupt transitions in both timbre and volume are minimised. Given enough samples (including round-robin ones), I'd take the natural dynamics any day.

The decades I spent programming transitions the way you described wouldn't have been necessary had I had the luxury of kazillions of samples on hand; crossfades wouldn't have been necessary either, FWIW, but I'm sure you'd agree with that as they're a right PITA where phase is concerned.

[David Polich]

This "normalizing" discussion is all a moot point. The samples in FHP may or may not have been normalized. But FHP is what it is. If you feel the level of something is too low, you'll just have to boost the output of Kontakt, either with a trim plug, or increasing the bus level for the audio track you're recording the part to.

I've already used FHP in a production, and I can tell you the level of instruments is NOT a problem.

And I'm certainly not going to go in at the microscopic level and edit individual WAV samples
in a Kontakt instrument. Life is short, I don't have the time nor the inclination. But even if I were to undertake such a task, I would never normalize anything, ever. I'd use my audio editor's "Change gain" on a particular sample to make it louder. Normalizing always brings up the noise floor. It is a holdover from the bad old early days of digital, before "change gain" became a feature of audio editors.

Anyone is free to contact the authors of a particular library and make suggestions on how to process their sample data.

[stubbsonic]

In the last option, riding levels, do you hear any trouble with background noise as you crescendo?

Regarding noise, as you hinted at, Shooshie, you are not adding any. You are simply boosting both the signal and the noise that were present to begin with. Whether you boost levels in recording or through normalizing, you are giving yourself the opportunity to address the noise in more effective ways. But NR isn't always necessary, as the quiet/noisier samples will be played back quietly anyway-- just as they are in the STAIR-STEPPED dynamics version.

The only difference I can see, apart from possible artefacts introduced by the normalisation process,

I don't hear artifacts added by normalizing (that's not to say it doesn't happen, I just don't have those kind of ears, I guess.) Normalizing is, after all, just gain. Do we introduce artifacts every time we use a fader? If you are using "artifacts" in a very broad sense to mean the low-resolution of the low-level recording, you aren't adding them, you are just boosting them proportionally to the level of the signal.

I thought that the angle was that instead of introducing artificial dynamics (mapping normalized files to velocity), the samples are instead mapped directly to velocity, thus preserving the natural progression as played by the drummer.

But you are ignoring the intent of the end-user as player, which is arguably more important.

Yes, I'll concede that in Option 1 you DO preserve the original relative volumes of the range of hits. You preserve them as stair-steps of volume. And how important that is, is perhaps the crux of this conversation. But we have to consider the response of the instrument to the end-user. Because in ANY case, the programmer is artificially assigning a velocity-range to a sample.

In Option 1 (used by HHS), the programmer may have assigned velocities 75-90 to play a sample from layer 7. However, all 16 of those input velocity values will play that sample at the same, fixed output level. You could say that the mid-point was the arbitrary velocity-to-level that was chosen by the programmer. But you will be farthest from the end-user's intent at 75 and at 90.

... the artificial mapping (formerly-mentioned method) is a process borne out of theoretical perfection, that is, an even transition envisaged by the programmer, whereas the latter directly links the dynamics played to velocity, human variations intact.

But this assumes that some minor deviation in volume from the original relative level will be more obtrusive than stair-stepped velocity response.

In other words, the volumes of the relative dynamic samples may not increase in a perfectly-linear fashion in real life, whereas when normalised, mapped to velocity and controlled by the VI / synth's amp engine, said levels would surely follow a more smooth path from quiet to loud.

First of all, the end-user isn't going to be able to control the velocities of the input device-- be it MIDI keys or drum-pad, so the human factor is not to be lost in Options 3/4. Secondly, if the end-users bangs within the range of a single layer, there will be NO volume variation (except that introduced by the round-robin samples). But ultimately you would not lose the human response in any case.

The thing is, that, using your example, two different strike-levels' being represented by the same sample, albeit with different amp levels, isn't realistic. In reality you'd hear differing timbres between those two strikes. I'm guessing many more than 10 layers have been employed, ensuring abrupt transitions in both timbre and volume are minimised. Given enough samples (including round-robin ones), I'd take the natural dynamics any day.

I'll concede that with enough velocity switched layers and round-robin samples, you won't notice the instrument ignoring your velocity choices as an end-user. So let's ask the obvious question. How many velocity layers are enough?

The decades I spent programming transitions the way you described wouldn't have been necessary had I had the luxury of kazillions of samples on hand; crossfades wouldn't have been necessary either, FWIW, but I'm sure you'd agree with that as they're a right PITA where phase is concerned.

I rarely used cross-fades for obvious reasons. Though when I had a bad transition, I sometimes made a "hybrid" using frankenstein techniques.

[monkey man]

The only difference I can see, apart from possible artefacts introduced by the normalisation process,

I don't hear artifacts added by normalizing (that's not to say it doesn't happen, I just don't have those kind of ears, I guess.) Normalizing is, after all, just gain. Do we introduce artifacts every time we use a fader? If you are using "artifacts" in a very broad sense to mean the low-resolution of the low-level recording, you aren't adding them, you are just boosting them proportionally to the level of the signal.

I don't either. I included this qualified statement (note use of word "possible") to cover those who'd use it as an argument. Had anyone done so, I simply wouldn't have gone there.

I thought that the angle was that instead of introducing artificial dynamics (mapping normalized files to velocity), the samples are instead mapped directly to velocity, thus preserving the natural progression as played by the drummer.

But you are ignoring the intent of the end-user as player, which is arguably more important.

Yes, I'll concede that in Option 1 you DO preserve the original relative volumes of the range of hits. You preserve them as stair-steps of volume. And how important that is, is perhaps the crux of this conversation. But we have to consider the response of the instrument to the end-user. Because in ANY case, the programmer is artificially assigning a velocity-range to a sample.

In Option 1 (used by HHS), the programmer may have assigned velocities 75-90 to play a sample from layer 7. However, all 16 of those input velocity values will play that sample at the same, fixed output level. You could say that the mid-point was the arbitrary velocity-to-level that was chosen by the programmer. But you will be farthest from the end-user's intent at 75 and at 90.

I assumed a great number of samples, as this seems to have become the norm these days, what with fast HDDs and so on. You assume 10 or less where I assume perhaps 64 to 127, one for every MIDI velocity increment or two. You're correct given your assumption, and I, dare I say it, am correct given mine.

... the artificial mapping (formerly-mentioned method) is a process borne out of theoretical perfection, that is, an even transition envisaged by the programmer, whereas the latter directly links the dynamics played to velocity, human variations intact.

But this assumes that some minor deviation in volume from the original relative level will be more obtrusive than stair-stepped velocity response.

If you're talking whopping-great steps, which, given your limited-sample count assumption, you are, then you're correct. I'm saying that, even if one had 127 samples, one for each MIDI-note velocity level, the transitions in volume may not be mathematically equidistant in reality, whereas synth-engine-imposed ones are.

I agree with David, and at this level, I suspect you do too; it's ridiculously subtle at any rate.

In other words, the volumes of the relative dynamic samples may not increase in a perfectly-linear fashion in real life, whereas when normalised, mapped to velocity and controlled by the VI / synth's amp engine, said levels would surely follow a more smooth path from quiet to loud.

First of all, the end-user isn't going to be able to control the velocities of the input device-- be it MIDI keys or drum-pad, so the human factor is not to be lost in Options 3/4. Secondly, if the end-users bangs within the range of a single layer, there will be NO volume variation (except that introduced by the round-robin samples). But ultimately you would not lose the human response in any case.

Again, I agree where a limited sample count is employed. I've no idea how many FHP uses, but I'll bet it's a heckavalot more than, say, 10.

The thing is, that, using your example, two different strike-levels' being represented by the same sample, albeit with different amp levels, isn't realistic. In reality you'd hear differing timbres between those two strikes. I'm guessing many more than 10 layers have been employed, ensuring abrupt transitions in both timbre and volume are minimised. Given enough samples (including round-robin ones), I'd take the natural dynamics any day.

I'll concede that with enough velocity switched layers and round-robin samples, you won't notice the instrument ignoring your velocity choices as an end-user. So let's ask the obvious question. How many velocity layers are enough?

Obviously the answer is one for each possible velocity amount. In the case of MIDI, excluding "0", that'd be 127.

Highlighted type: With "enough velocity-switched layers and round-robin samples", you won't notice this at all, because the instrument isn't ignoring your choices; it's merely mapping velocity levels to samples, and with enough sample layers at play, there'd be nothing to notice. I suspect you'd imagined many more layers in this example, but not quite enough to reach the point where ranges of velocities didn't call up single samples (ignoring round robins for the purpose of the argument).

The decades I spent programming transitions the way you described wouldn't have been necessary had I had the luxury of kazillions of samples on hand; crossfades wouldn't have been necessary either, FWIW, but I'm sure you'd agree with that as they're a right PITA where phase is concerned.

I rarely used cross-fades for obvious reasons. Though when I had a bad transition, I sometimes made a "hybrid" using frankenstein techniques.

Frankenstein techniques. LOL

That was my point. They'd not have been necessary as transitions wouldn't have been obvious had we old fogies had more samples to work with.

I'm done with this. I'll leave it to those who get paid for it to sweat the small stuff, as I'm confident the folks at HHS have done for years. I feel your concerns would've been front and centre when they put the library together; it is, after all, arguably the principal stumbling-block when creating this sort of thing.

Take care, Stubbsy. Hey, that's what I used to call you; it's the Aussie way, mate. Hope you don't mind.

[stubbsonic]

Seems like we are in basic agreement.

But I'm less optimistic about how many velocity layers there are in many of these libraries.

Another point is that it is VERY difficult to get a drummer to interpret more than a handful of dynamic levels with any amount of control.

Can you ask your sampling session drummer to give you 50 hits, gradually increasing from ppp to fff? Then do you do that for every kind of hit? Then keep track of those, plus have round-robin?

I'm guessing it is more realistic to have 10 velocity layers.

EDIT: Oh, and I have a dear friend who calls me "stubbsy". Well, now I have two.

[monkey man]

Seems like we are in basic agreement.

Agreed.

Another point is that it is VERY difficult to get a drummer to interpret more than a handful of dynamic levels with any amount of control.

Can you ask your sampling session drummer to give you 50 hits, gradually increasing from ppp to fff? Then do you do that for every kind of hit? Then keep track of those, plus have round-robin?

I've thought about that long and hard over the years. I'd be surprised if machines weren't used for this sort of thing these days.

I remember many years ago someone did it (I can't remember who), so you'd think folks would've taken up the baton and run with it, if indeed it yielded satisfactory results.

I'm guessing it is more realistic to have 10 velocity layers.

Well, the drummer, assuming a machine isn't employed, could execute a shedload of hits and they could easily be reordered in a DAW. In theory it should work and I reckon this'd be the thing to do if a mechanical option weren't available.

EDIT: Oh, and I have a dear friend who calls me "stubbsy". Well, now I have two.

Stubbsy it is then. Yay!

Cheers bud.

[stubbsonic]

Just to follow up, I was looking at some of their freebie sample sets.

There were anywhere from 4 to 12 velocity levels. Right hand and left hand hits, as many as 4 round-robin hits. So ONE very simple sound set hat 970 samples, and 8 velocity levels was pretty typical.

The random variation in level from the round robin samples does obscure the stair-steps in level. But just wanted to show you that the dynamic resolution is pretty limited by this approach. Is it a problem in regular use? Probably not. There are times when I do kind of fast rolls with LOTS of dynamics. And I can hear and feel it when it doesn't react properly.

Since MonkeyMan was suggesting they have more velocity layers than that, I just wanted to give that example.

[jloeb]

I'm certain that there's some valence to this technical discussion.

Just want to unambiguously state that, to me, an experienced percussion player (and many others, some of whom have published reviews), the sound quality of the FHP library is absolutely unimpeachable, and one to which other libraries should (and have) aspire/d.

[monkey man]

Since MonkeyMan was suggesting they have more velocity layers than that, I just wanted to give that example.

Just speculation, Stubbsy; there was no way I could know.

We both knew that RRs could obscure the transitions, and it's good to know this seems to have turned out to be the case.

All in all, sounds like one can't possibly go too far wrong with this set. I'm certainly in... when that first plug purchase comes 'round in a few months or whatever. Sounds exciting, especially given that I've never owned a decent library.