The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Should you Dither when converting your 24Bit sample libraries to 16Bit? Drum Machines & Samplers
Old 21st December 2010
  #1
Gear Addict
 

Should you Dither when converting your 24Bit sample libraries to 16Bit?

Also, while I'm on the topic... Is it a good Idea to re-normalise everything to 0DB while doing this?

My guess is Yes for both questions...but only use Pow-r #1?
Any thoughts?

Cheers.
Old 21st December 2010
  #2
Lives for gear
 

Dither: yes
Noise Shaping: no

...never normalize to 0dB, I would go to -1dB at an absolute maximum, but even then you're risking "intersample peaks" of greater than 0dB...

You've still got just under 96dB of s/n ratio with a 16-bit file, there's no need to be peaking at 0...
Old 21st December 2010
  #3
Lives for gear
 
elambo's Avatar
If this was my project I wouldn't touch volume at all.
Old 22nd December 2010
  #4
Lives for gear
 

neither would I
Old 22nd December 2010
  #5
Gear Addict
 

Hmmm, I'm not so sure, I mean aren't Samplers designed to be able to work with Normalised files? ...I'm quite sure most implement sufficient headroom within the playback system...but maybe I'm wrong?

Ah yes, though it's for use in an MPC, so I want some sort of consistency in the volumes of the different sounds for performance reasons...

I still think I should use POW-r Dithering in the process...
Old 22nd December 2010
  #6
Lives for gear
 
elambo's Avatar
Quote:
Originally Posted by emilision View Post
Hmmm, I'm not so sure, I mean aren't Samplers designed to be able to work with Normalised files?
Of course, but that's not the issue. Unnecessary normalization is a bad habit for the integrity of the audio, and it might also make a mess of the dynamics of the samples.
Old 22nd December 2010
  #7
Lives for gear
 

Quote:
Originally Posted by elambo View Post
it might also make a mess of the dynamics of the samples.
This ^^^......having the same peak sample value can be very very different from being the same perceived volume.
Old 22nd December 2010
  #8
Gear Addict
 

Thanks guys for your answers BTW.
'Unnecessary normalization is a bad habit for the integrity of the audio, and it might also make a mess of the dynamics of the samples.'
This is one of those things that I've often heard mastering engineers go on about, but I've never really fully quite understood why...can you recommend any articles?
'...having the same peak sample value can be very very different from being the same perceived volume.'
Of course. However, my theory in this case is that: after a cursory glance through some of the samples (some of which are in fact 32Bit) I noticed the peak levels to be actually quite low in many cases, and so in converting such samples to 16Bit we would be losing quite a lot of information. Besides having to possibly adjust gains quite significantly in the programs. Normalisation seemed like a easy "rough fix" ...rather than a laborious picking through things individually.
Old 27th December 2010
  #9
Gear Addict
 

Hey again...

So, which one should I use... POW-r #1?

Cheers.
Attached Thumbnails
Should you Dither when converting your 24Bit sample libraries to 16Bit?-screen-shot-2010-12-22-20.41.19.png  
Old 27th December 2010
  #10
Lives for gear
 

I would use triangle (TPDF).
Old 27th December 2010
  #11
Lives for gear
 
jupiter8's Avatar
 

There are too many people who take this never normalize way too serious. Yes it can lead to bad results if you don't know what you're doing but the opposite is also true. There are times when it is justified to normalize. Just avoiding it because people say you should never do it is ridiculous. Learn the facts and do it when it benefits you and avoid it when it doesn't.

I mean what on earth does intersample peaks has to do with this scenario ? None.

It is funny how you should always take the outmost care of sound quality and there's no lengths you shouldn't go to, to get the best results but when normalizing comes up it's always evil and should always be avoided. Then 16 bits suddenly is way more than enough even though you would never submit your regular audio to such a treatment.
Old 27th December 2010
  #12
Lives for gear
 

In this case, the more valid point was made by Elambo in that normalising to a global value might mess up the volume interaction of the samples.

But you're right, I was being inconsistent.
Old 30th December 2010
  #13
Gear Addict
 

OK, now I'm even more confused...
Old 30th December 2010
  #14
Gear Guru
 
UnderTow's Avatar
Quote:
Originally Posted by jupiter8 View Post
There are times when it is justified to normalize.
When?

Samples peaking at 0 dB will have to be brought down in level during mixing anyway or your mix will end up over 0 dB FS. So why not leave them at a reasonable level to start with?

In the days of 12 bit samplers it made sense to normalise and worry about noise floors. With 16 bit it doesn't really make much sense any more. (With 24 bits even less).

As for the dither... You would be safe with TPDF (triangular) or you could use something with slight noise shaping, nothing too extreme.

Alistair
Old 31st December 2010
  #15
Lives for gear
 
elambo's Avatar
Quote:
Originally Posted by UnderTow View Post
When?
Almost never.

Quote:
Originally Posted by UnderTow View Post
Samples peaking at 0 dB will have to be brought down in level during mixing anyway or your mix will end up over 0 dB FS. So why not leave them at a reasonable level to start with?
Yep!

Quote:
Originally Posted by UnderTow View Post
In the days of 12 bit samplers it made sense to normalise and worry about noise floors. With 16 bit it doesn't really make much sense any more. (With 24 bits even less).
You're making too much sense. heh
Old 31st December 2010
  #16
Lives for gear
 
elambo's Avatar
Quote:
Originally Posted by jupiter8 View Post
Just avoiding it because people say you should never do it is ridiculous.
You're right, that WOULD be, so it's good that we've taken some time to explain, because it's equally ridiculous to avoid something simply because someone said you should.
Old 31st December 2010
  #17
Lives for gear
 
elambo's Avatar
Quote:
Originally Posted by emilision View Post
Hey again...

So, which one should I use... POW-r #1?

Cheers.
Hadn't meant to avoid the question - I've never tested them. I've heard preferences for 1 by some people and 2 for others. I assume it depends on the source material. Try a few and use whichever sounds best. Pow-r has nice algorithms so you're in good hands whatever you do.
Old 31st December 2010
  #18
Lives for gear
 
cdog's Avatar
Quote:
Originally Posted by emilision View Post
I noticed the peak levels to be actually quite low in many cases, and so in converting such samples to 16Bit we would be losing quite a lot of information. Besides having to possibly adjust gains quite significantly in the programs. Normalisation seemed like a easy "rough fix" ...rather than a laborious picking through things individually.
Guys, we are making music here, not mastering a jazz album.

Im gonna go out on a limb here an guess that the samples in question arent brush strokes on a snare drum or a mellow little flute part, but probably some kick hit off a 70s soul album which has already been compressed, limited and faceraped by the RIAA equalization curve, which just needs to be brought up to a proper level.This isnt some pristine garden of audio sample nirvana we are dealing with, most likely.

Normalizing a batch of sound files with wildly differing levels will allow them to all play together nicely in your sampler. Yes, an MPC and most if not all samplers are meant to work with normalized files. If you are worried, then fine normalize to -1 or -1.5, it wont make a practical difference really.

Normalizing a low level 24 bit audio file before converting to 16 bit is a standard process for uploading to a sampler, and is an extremely useful and practical process, although it isnt always desireable in every case, it often is, so dont let people tell you otherwise.

Old 31st December 2010
  #19
Gear Head
 

Quote:
Originally Posted by cdog View Post
faceraped by the RIAA equalization curve
I will have to remember that and use it somewhere, it is THAT eloquent...
Old 31st December 2010
  #20
Gear Head
 

Quote:
Originally Posted by timlloyd View Post
In this case, the more valid point was made by Elambo in that normalising to a global value might mess up the volume interaction of the samples.

But you're right, I was being inconsistent.
Exactly. I have actually had this issue come up when re-sampling a drum machine kit. If I normalized all individual samples, then the hi-hat or cabasa would be at the same volume as a booming kick drum, which would create a very robotic, unrealistic and annoying sounding drum track. This would ruin the dynamic range of the drum kit. The only solution would be to trigger EACH instance of a hi-hat sample at lower note velocity, which would just add extra tedious MIDI track and note editing. Or, when building the kit in a sampler, the volume of the individual hi-hat sample would have to be lowered.

If you are creating sample sets with multiple velocity layers, normalizing will effectively cancel the volume AND sonic characteristic differences between the layers.

What I did do, though, was record each sample in the kit consecutively on one audio track, and normalized the OVERALL track, effectively maintaining the volume difference between individual instruments.

Since samples will usually be used in a project, not mastered for a CD by themselves, there will be additional effects and processing added to them when they are used in a VST instrument. To allow headroom for this processing later down the road, I normalize samples to -6dB. I stay in the 24-bit/96kHz domain all the way to the very end, when the final project gets mastered for 16/44.1. By the way, the mastering engineer also appreciates the 24/96 and -6dB consideration. It makes his or her job much easier, too.
Old 31st December 2010
  #21
Lives for gear
 
elambo's Avatar
If someone absolutely insists to go against conventional wisdow, the proper way to normalize a grouped collection of samples (e.g. various velocity layers for one note one a piano, or a snare, etc.) is to assess the gain required to raise the loudest sample - the top velocity - to -0dbFS, then apply the same amount of gain to every other sample in the group. So if your top layer is -6, add 6db to that sample and 6db to all others as well. This would be a poor practice because you're left with absolutely no headroom, but at least it would maintain relativity for that instrument.
Old 31st December 2010
  #22
Lives for gear
 
jupiter8's Avatar
 

Quote:
Originally Posted by UnderTow View Post
When?

Samples peaking at 0 dB will have to be brought down in level during mixing anyway or your mix will end up over 0 dB FS. So why not leave them at a reasonable level to start with?

In the days of 12 bit samplers it made sense to normalise and worry about noise floors. With 16 bit it doesn't really make much sense any more. (With 24 bits even less).
It is funny how sometimes DSD isn't even a good enough medium but when it comes to samples people have no qualms about going 12 bit or even less.
Yes i know it can be done for an effect but if you want the highest fidelity out of a 16 bit sample you need to normalize it (coming from a 24 bit sample that is)

People seem to think that normalizing is utterly destructible and will ruin a good sample when it is nothing even remotely that.

You can for maximum quality/size ratio record a sample at sensible levels (say -18 dB) at 24 bit,normalize it and convert to 16 bits. That way you have maximum fidelity and minimum size. If you're interested in maximum fidelity that is the most efficient way to go. But if you can't be bothered to move a fader i guess not normalizing makes sense too.....

Quote:
Originally Posted by elambo View Post
You're right, that WOULD be, so it's good that we've taken some time to explain, because it's equally ridiculous to avoid something simply because someone said you should.
If you don't know that may i suggest you don't advice people on whether to normalize or not.
Old 31st December 2010
  #23
Gear Guru
 
Muser's Avatar
it's complicated by the application. if I drop some samples to -5db peak and put them into a Akai Z8 the filter doesn't grip them well.
If I up them by 1db it does.

so for my application -3db seems a good value for a Z8.

however, if the sample has a lot of high frequency content and I intend to put it in a Roland MV8K then I need to drop them to -13db, because after I pre-emphasize them in the Roland, they are ruined. the MV has no gain change algorithm to do it internally.

if you are using velocity dependent layers in a patch, that's actually when you want the samples to be at the same peak value.
otherwise the midi velocity will extend the dynamic range along with the sample also having differing db scales.
so it won't work very well to have different peak levels in samples which are bound for a velocity response group.
Old 31st December 2010
  #24
Lives for gear
 
elambo's Avatar
Quote:
Originally Posted by jupiter8 View Post
If you don't know that may i suggest you don't advice people on whether to normalize or not.
I welcome you or anyone else to poke a hole in any theory I've written in this thread. I'm not out to win some prize for being correct, I'm out to try to help someone who'd asked a specific question about a subject I understand. If I'm wrong, inform me -- I have the capacity to reconsider my stance on normalizing, in fact it would be to my benefit. So far I've read nothing that has me even remotely curious about normalizing samples. Some good theories, and certainly some good intents, but they're easily shown to be impractical in modern technology.
Old 31st December 2010
  #25
Lives for gear
 
elambo's Avatar
Quote:
Originally Posted by Muser View Post
it's complicated by the application. if I drop some samples to -5db peak and put them into a Akai Z8 the filter doesn't grip them well.
Z8 filters are sensitive, but adjustable, no?
Old 31st December 2010
  #26
Gear Guru
 
Muser's Avatar
Quote:
Originally Posted by elambo View Post
Z8 filters are sensitive, but adjustable, no?
yes they are. but I think the sample first feeds the filters. they also have attenuation in the filters down to -30bd. but I'm not sure if that attenuates only the resonance in some way. the gain staging in the Z8 is pretty odd to work out. but in the specific case I tested it in, the difference in 1bd was obvious to some performance in the filter.

so, something must be the case in other cases as well with the Z8 regarding the the initial sample db.

I think I actually tested the attenuation against 3 of the same samples. one @ -6db one @ -5db and one @ -4db.
The difference in this case between -5db and -4db was quite startling. I don't think the Z8 attenuation control made any big difference
to the way the filter initially gripped the sample. but the one db difference in the sample did.
the attenuation goes in minus 6db steps anyhow. so there would be a lot to fully test there.
Old 1st January 2011
  #27
Lives for gear
More dithering info:
Logic Pro Help :: View topic - Logic's Dither Questions

POW-R - Wikipedia, the free encyclopedia

Wiki mentions PowR#1 was designed for speech, PowR#2 Dynamic range of rock music, PowR#3 wider dynamic range/ Orchestral type music
Top Mentioned Products
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Similar Threads
Thread
Thread Starter / Forum
Replies
zzz / Mastering forum
3
FeatheredSerpent / So much gear, so little time
9
TacosWhenTwisted / Music Computers
11
duvalle / Mastering forum
2
doug_hti / So much gear, so little time
9

Forum Jump
Forum Jump