EDIT: So I'm a little bit in now and have part of the demo so you guys can see what direction I'm going in. Just a note, the first track until the little breakdown bit is without transitions to highlight how much difference they make (any comments welcome!):
http://eternitysound.com/mp3s/Chemical_Kinetics_Take_1.mp3

EDIT: Initial woffle woffle about samples rates and editing removed...
Last edited by GeorgeZ on Fri Jun 22, 2012 1:15 am; edited 5 times in total

24 bit

I usually use 24bit 44k wav ...

I release almost everything in 48 Khz/24 Bit - the alternative beeing 44.1 Khz/16 Bit for less dynamic material like e.g. Drumloops.

What as been said so far should be enough

Anyway just to explain a bit: 24 bit is the ideal and you should stick there but, as Sampleconstruct noted, if you have material with low dynamic range you can recover a lot of Mb going 16 bit without quality loss.
For what concerns frequency the loss of quality between 44 and 48 is less than a sample rate conversion loss hence I usually think in this way: if the material it's aimed to audio production I go 44.1k otherwise if video production is my target I use 48k.
If in doubt is generally better 44.1k 24 bit than 48k 16bit occupying almost the same disk space.

16 bits or 24 bits is alright. Maybe it might be worth it to offer both? Sometimes, depending on the type of sounds, I prefer to stick with 16bits (for old drum machines samples, vintage films clips, etc...).

Cheers lads! Well, it's VERY dynamic material, but I'm doubtful that my humble little pack will reach the movie/video industry, so 44.1k 24 bit it is... follow on question: seeing as I've done the bulk of it in 32bit-float, is it ok to batch convert to 24bit? Any suggestions for a reliable program to do it with?

Convert or no?

Forget about 32-bit. This is completely unnecessary and it isn't true 32-bit.

As has been said before, 24-bit/48Khz is probably the best way forward.

Creating a sample library, even small ones which you may even be giving away for free, need planning. Here's my top tips:

1) Sample something relevant and meaningful.
2) Consistency of sound quality and levels are highly important.
3) Samples must always be topped and tailed effectively.
4) Provide multiple velocity layers if necessary. This can be good even in synth sampling, where a higher velocity gives a tiny difference in sound or some kind of modulation happens. In this case, it is sometimes better to create the sample format (say HaLion and/or Kontakt)
5) Selling samples is hard. VERY hard. Trust me!! Consider giving (some of) your libraries away for free. That way people will see the quality you offer and will be more likely to part with their hard earned cash on your future releases. Musicians love a free meal - they will thank you for it, and the best ones will return the favor!

Happy sampling!!

Hi James,

Thanks for the tips, definitely some useful ones in there.

I'm up against something else now... not having dealt with sample libs before and not using them myself much I'm wondering, what's the average amount of samples in a lib should be? I realize this can vary wildly depending on the content and there probably isn't a set answer, but lets just say it'll be a sub $10 lib (I want it to be cheap as chips and a bargain for the amount of content, not looking to rip anyone off ) with ascending, descending and sweep transitions.

Going by the Vengeance standard of pricing (which is a relative average figure of course, and they are VERY known) that would be $20 for 150 samples... that seems awfully expensive to me! What do you think?

GeorgeZ, not sure where you ended with the conversions question, but as mentioend you want to end up at 44.1k/24b or 48k/24b. Practially speaking, you can probably get away with 16b with the proper dither, etc., but for the sake of saving on some superfluous email correspondence and angry KVR threads, go with 24b from the outset.

Do do do go with Voxengo r8brain for all your batch conversion needs, unless you already have a WAV editor that can do what you need. The SRC from r8brain was in the top-tier not too many years ago, though most DAW and sample-based software has "caught up" to it. Still a useful tool for sample manipulators, however. On the bit reduction side it gives you a flat dither (i.e., no noise shaping), which is what you want, and of course batch capability. The one app gives you a single place to do all your SRC and bit conversions, which is nice, if that's what you're, you know, looking for.

Oh yeah, the freebie works just like the Pro version, but is limited to linear-phase SRC only at 48kHz max output -- i.e., no minimum-phase SRC, and while you can convert higher kHz down to lower kHz, your final output kHz can't exceed 48kHz. Most people wouldn't need more than the freebie.

Free: http://www.voxengo.com/product/r8brain/
Pro: http://www.voxengo.com/product/r8brainpro/

For some discussion on SRC, and how it can screw up your life see here, for example: http://src.infinitewave.ca/ and http://src.infinitewave.ca/help.html

Interesting to compare the SRC of some of those older app versions, etc., and just how crappy it was.

In fact, with the above comparison site you can compare r8brain lin-phase and min-phase output... not too shabby for freeeeeeeeee!

EDIT: for bit reduction, any time you reduce effective bit resolution of a sample you should dither. In practice, if you went 32-bit float down to 24-bit int and just truncated the missing bits rather than dithered, 99.9% of people wouldn't catch it. In any case, just make sure you dither even in the practically unnecessary case, 'cuz it's just the right thing to do. A tool like r8brain and others gives you the dither in the processing, sometimes with extended options for type of dither, etc...

Hi ya mate.

Thanks for that expansive explanation! Really appreciate time you put into it!
I've decided to go with 44.1k/24b. Now, I've tried to do a batch conversion with Waveosaur and that worked just fine. It also did a good job of reducing sample size (which is part of the reason I needed to downsample... 32bit float files are MASSIVE!). After your post naturally I was inclined to do the same with r8brain to test the results (also had a good read of the site and relevant links you posted, very interesting stuff... and Wavosaur's conversion looks pitiful on there ). I found the batch conversion of r8brain(free) a bit awkward. Had to make a copy of my original patches and directory structure as r8brain renames the pre-conversion files... not good when dealing with a 100+ sample lib. Thankfully I did make a copy to do the tests with Anyway, not a biggy, just something to take note of. So, did the conversion and all went well. After looking at the test results on http://src.infinitewave.ca/ I figured there'd possibly be a difference in audio quality between the Waveosaur and r8brain conversions. I tested the conversions vs each other and vs the original (32bit float @ 44.1k) via phase reversal. Before I get to that result though I think it's worth mentioning that Waveosaur actually reduced the file size by quite a bit, which is good of course, while r8brain did practically nothing to the file size (I though it would have at least reduced it by a bit) ...?

Anyway, so I did the phase reversal test and there was no difference in the sample quality (as far as I could tell). 100% nulling of the signal at reversal on all comparisons. So I'm a bit baffled. Is there any point in using r8brain then if there's a no discernible difference? Waveosaur is a tad easier to use, and the fact that I can't see any difference in the output, apart from the fact that the Waveosaur output file is a decent chunk smaller, I guess that would be the smart choice right? I mean, I'm not saying that those tests are incorrect, I just think they make a big fuss over something that is in essence actually a very small thing, at 44.1k/24bit in anyway. Are there that many people who will even be able to pick an infinitely small bit of spectral duplication past Nyquist in my samples? I'd venture a guess to say probably not

Maybe I'm not understanding the relevance completely, however the null test was pretty conclusive I think.

Thoughts?

Well, my one summary thought on your direct experience would be: if it sounds good, the charts and graphs don't really matter, do they?

Expanding a bit... those comparison graphs represent worst-case/best-case scenario, but not necessarily real-world results. Those technical comparisons are useful if you know ahead of time that for the material you're dealing with you need the absolute cleanest SRC output, and that not all SRC are created equal. And some can produce dead clean results, while others can produce dirty results. Whether the "dirty" manifests itself in any way, again, depends on the content itself, and the context of it's usage. I've certainly gotten great practical results in my FLStudio projects, SONAR, Reaper, etc. etc.

That being said, the one bit of practical advice I would give you in your case is to not only do your own qualitative tests, as you've done, but also think about how your samples will be used... did the conversion produce so much distortion/artifacts that when you push it through an enhancement pipeline -- substantial compression, eq, tubification, spectral whatever -- does that dirtiness become a real audible problem? In other words, just because you can't hear or see (much) anything in a null test doesn't mean that there isn't a significant difference for certain usages. The question is: is that difference ever going to be a problem for somebody down the road? E.g., if I want to run something through an aggressive Amplitube patch (not entirely unreasonable use case), do those ugly artifacts really come out in a bad way? Just try a few practical, real world setups, and listen with phones through decent A/D and amp to hear the fine detail... If it all sounds clean in various scenarios then it is clean, regardless of what a graph of a sine sweep may tell you. Know what I mean?

RE: r8brain overwrite... I could have sworn that even the free version created *_r8b.wav files, or whatever they are... or not overwriting is at least an option. Perhaps not an option in the free version? Not sure, but if so, that's a bit annoying. I have used the pro version myself for many years, but mostly for going 44.1 <=> 48 and the like transforms.

RE: SRC comparisons... I noticed last night on that infinitewave.ca site that the command line audio utility "SoX" was listed, and shows high-end SRC results for it. For sample developers SoX is a potentially very useful set of command line utilities for other things as well, like chopping up samples by threshold dB, changing sample format (WAV, ogg, aiff, mp3, etc. etc.)... I know I've used it for my own stuff quite a bit over the years when even a CoolEdit macro wasn't giving me warm and fuzzies. To quote from another thread where I provided a hypothetical SoX command line (Unix shell, to be exact) for a specific batch process:

http://www.kvraudio.com/forum/viewtopic.php?t=333118&highlig ht=sox


If you're of the Bare Metal inclination, SoX info is here: http://sox.sourceforge.net/

Been around for ages, and been a very useful toolset for just as long... Need to be familiar with Windows batch or Unix shell (Linux, OSX, Windows unix emulation like "cygwin", etc.) to get the most out of batch processing with SoX.

24bit/44.1k

A few things I've learned as a developer:

1. Low price does not equal more sales. As a matter of fact, I'm beginning to think that pros are wary of too low a price, and steer away from possibly wasting time (and money) on something that could be useless to them. Also, we have libraries priced at $5 and $40 and $50. We sell WAY more of the more expensive ones, and very few $5 libraries. Freebies may foster a bit of goodwill, but I don't believe that they generate many sales (we do offer some free libs). Many people will not even bother downloading a free library, since they believe it's a waste of time.
Still, in the beginning, it might be useful to get feedback.

2. Make your library easy to use! As a composer, I wouldn't buy a library that consisted of just wav files. It's just a waste of time to audition them. Your best bet is to organize them into keygroups that can be triggered easily from a keyboard. Kontakt is the standard, if you want to sell libraries. If you don't have Kontakt yet, at least build your instruments in a format that can be imported into Kontakt. To be fair, there are other sampler formats that do a fine job.

3.Make great and honest demos! Be careful not to give your product away by exposing your samples naked, but give your potential customers a clear idea of what your library can do! Your customer should be able to easily get the same sound as they heard on your demo. Create mp3s and more importantly video demos that show what the library does. Once again, you can see how important ease-of-use is. You don't want a video showing a cursor clicking on some loose wav files in iTunes!

Don't get too caught up in the technical choices. As long as you have clean audio (using a decent A/D interface- not the 1/8" mic input on the side of your laptop) you should be fine. A 16bit sample library will sound better than a 24bit one if the source sounds are better. Edit everything with care. No loose ends.

Making a sample library is hard work. Selling it isn't that hard if you create something that people want and have you have good demos. Best of success to you!