Capturing Samples from Software Instruments - how 'hot'?

[Liney]

This is a fairly basic question, apologies if it has been answered before (I did do a search of the forum).

I know a little about gain staging etc generally, but I feel like I have heard opposing views on this particular question: if you have designed a sound in a soft synth e.g. a kick drum, and you want to bounce it to audio to use in Kontakt, how hot should you be bouncing it?

Intuition tells me that, as long as it isn't clipping, it should be as loud as possible, in order to not waste any zeroes and ones - hence leave about -1.0dB head room?

But I feel as if I've read elsewhere to leave at least -10.0dB when capturing samples.

Is there a definitive correct answer?

[stardustmedia]

ITB created samples can be -1dBFS and using the whole headroom. I don't see a technical reason to go lower.

Maybe the final instrument that uses those samples, have some analog circuit emulation, thus hot samples will drive it good, leading to (wanted) distortions. On the other hand, highly likely those instruments feature a gain per sample, so...

There is no definitive ONE correct answer?
I'd decide the level, on how they will be used next.
Just one thing I'd avoid: Bounce at a very very low level.


It's another thing when recording analog sounds from analog gear, because -1dBFS (with a sine-like signal, e.g. a steady bass) will mean you're driving your analog side pretty hot (generally converters are biased +4dBu = 0 dBVU = -18dBFS - measured with a voltage of 1.228V). If it's the sound you're looking for, just drive it. If you want to have a "cleaner" analog sound, without saturation/distortion/drive, you'll need to be around 0dBVU = +4dBu (again, for a sinus-like signals, they deliver a steady RMS). For sounds with strong transients, e.g. hihat, you won't be able to reach 0dBVU without analog drive/saturation. Probably the VU-meter won't even move, while already having a good gain, it's a "loudness" meter, not a peak meter.

[Liney]

Thanks for the reply - I am talking exclusively about recording ITB instuments, bouncing them within Logic. So it sounds as if I don't need to leave any headroom, as long as there is no digital clipping?

[stardustmedia]

Nope, in most cases no need to leave headroom. Of course, please no digital clipping
Unless even that can be desired in special circumstances

[aciddose]

If you're sampling in native floating point, the gain makes absolutely no difference.

If you're sampling in a fixed integer format like 16-bit or 24-bit, the question is: why? There are only negatives associated with this method and it simply shouldn't be done. If you need to convert to some integer format that can be accomplished afterward in a way that minimizes noise, often by normalizing all the samples before converting them to the format with less accuracy.

In floating point (the native signal from plug-ins) you can have a level anywhere between about -200 dB and +200 dB without worrying about losing any bits at all.

So it doesn't matter if you sample at +90 dB or -180 dB.

[stardustmedia]

If you're sampling in native floating point, the gain makes absolutely no difference.

If you're sampling in a fixed integer format like 16-bit or 24-bit, the question is: why? There are only negatives associated with this method and it simply shouldn't be done. If you need to convert to some integer format that can be accomplished afterward in a way that minimizes noise, often by normalizing all the samples before converting them to the format with less accuracy.

In floating point (the native signal from plug-ins) you can have a level anywhere between about -200 dB and +200 dB without worrying about losing any bits at all.

So it doesn't matter if you sample at +90 dB or -180 dB.

If the instrument where the samples end support such a format, yes. Otherwise, go for 24bit and aim for peak @ -1dBFS. Nothing to worry about loosing any bits.

I wonder what negatives is associated with 24bits?
16bit definitely not, unless, again, your destination only support 16bit.

[Liney]

Never heard of Native Floating Point, but my DAW is Logic Pro X, and I'm creating sounds in Absynth, running them through effects, and then bouncing them for use in Kontakt.

[aciddose]

Your DAW supports bouncing to float (perhaps a plug-in is needed) and Kontakt almost definitely does support loading float samples too.

The problem with bouncing to 24-bit or 16-bit is you lose a lot of precision. Why would you rip half the shingles off your roof and toss them in the garbage? "I don't really need those anyway, it'll be fine without them." Or maybe trudge around bare-foot "shoes aren't really that important when you have thick callus pads on your feet."

You need to do things like normalize, dither and so on to ensure you get maximum precision from 24-bit. It's extra "make work" that you don't need to do at all if you use float instead. It's just a waste of time unless you need to save disk space, memory, bandwidth (you actually only save 25% at best, often none at all) or you're using devices/software that aren't compatible.

A really simple explanation is: it's intentionally mixing a bunch of extra noise into the sample. Why would you want to do that?

I called it "native" floating point because most of the major plug-in standards use floating point data to transmit signals between plug-ins and the host. So no matter what you do, the signal is going to start out as float, get stored in some format, loaded by your sampler and then converted back into float again before it gets output by the plug-in. So it makes sense to skip the in-between conversion steps and just keep the signal in that native format in a lot of cases.

[stardustmedia]

Your DAW supports bouncing to float (perhaps a plug-in is needed) and Kontakt almost definitely does support loading float samples too.

The problem with bouncing to 24-bit or 16-bit is you lose a lot of precision. Why would you rip half the shingles off your roof and toss them in the garbage? "I don't really need those anyway, it'll be fine without them." Or maybe trudge around bare-foot "shoes aren't really that important when you have thick callus pads on your feet."

You need to do things like normalize, dither and so on to ensure you get maximum precision from 24-bit. It's extra "make work" that you don't need to do at all if you use float instead. It's just a waste of time unless you need to save disk space, memory, bandwidth (you actually only save 25% at best, often none at all) or you're using devices/software that aren't compatible.

A really simple explanation is: it's intentionally mixing a bunch of extra noise into the sample. Why would you want to do that?

I called it "native" floating point because most of the major plug-in standards use floating point data to transmit signals between plug-ins and the host. So no matter what you do, the signal is going to start out as float, get stored in some format, loaded by your sampler and then converted back into float again before it gets output by the plug-in. So it makes sense to skip the in-between conversion steps and just keep the signal in that native format in a lot of cases.

Logic does not support bounce to 32 bit.

To the OP: 24bit fine, do not bother yourself with the mentioned precision loss. That is theoretically true, but so low, that in almost all situations, it doesn't matter.

[aciddose]

It's more about what makes sense, and what is only a waste of time/effort.

If the host doesn't support multiple formats that's an unfortunate side-effect of exactly what you're saying "it doesn't matter", well, except when it does, and then if you're a Logic user you're out of luck eh?

[funky lime]

The problem with bouncing to 24-bit or 16-bit is you lose a lot of precision. Why would you rip half the shingles off your roof and toss them in the garbage? "I don't really need those anyway, it'll be fine without them." Or maybe trudge around bare-foot "shoes aren't really that important when you have thick callus pads on your feet."

A whole mix, perhaps. But if one of your (sampled) synth tracks is firing off at 24 bit, while the other 15 tracks are in float, who can even tell?

Float vs 24-bit is not quite as extreme as shoes vs no shoes. A more apt comparison would be wearing $500 shoes vs. wearing $200 shoes. Either way, you've got some really rather nice shoes on your feet.

[aciddose]

No, using a non-native signal format like 24-bit is like hand-made shoes that you stitched yourself. It takes a hell of a lot of extra work, is always worse than the mass-produced product (native float) and is generally just a waste of time.

You can buy decent quality sneakers for $50 if you just want to cover your feet.

Also, how much is that in libraries of congress?

[funky lime]

Ok sorry