pouët.net

So,

I'm scratching the good old Protracker itch. AGAIN. Means: I'm preparing a track in Bitwig in the 16bit domain, then export samples and sounds for conversion to 8bit.

The loss of information also seems to affect the level as well as dynamics. Which is kinda obvious. I wish Bitwig, or any other DAW, would be able to work in 8bit so I'd be able to natively work with the true levels and dynamics. Since that's not possible, and I can't be the first person to run into this issue:

What are your processes like? Just slam the 8bit samples into clipping and let that solve the problems? Any further editing, EQing, limiting you're applying? If so, how would you do that if modern editing tools don't allow for processing 8bit samples?

And, what's really getting lost during conversion from 16 to 8 bit?

Am I being too anally retentive? hmhmhm.

Enlighten me!

for drums: Usually clip them in PT. I have had nice results with fex. (EQ)->Omega 458a->(EQ)->Your Favourite Smashing Compressor Retaining Peaks Solution (may have more than one compressor & processor)->(EQ)->PT Gain clipping.

for other sounds: I try to keep high gain in final stage when importing to protracker and use few samples around the needed notes, may delete later on and just use 1 or 2. EQ obviously, may want to sometimes exaggerate stuff and let Paula handle it crispy 8)

So I guess I usually dont leave much dynamics left in the end at least when wanting something banging, but the original wave usually has exaggerated transient information.

I also wonder what people do. I've heard incredible sound on 16/32 bits machines and modern demos, but I'm not sure it's only due to the quality of the conversion. Sure it should add something to the tune. I know Izotope Ozone 8 have that dither thing with 8 bit bit depth and you can set the amount of noise shaping and dither. Maybe some free tools does it as well.

I just apply some compression and eq in Cubase and save the samples as 8bit mono wav. After importing them into the protracker clone I enjoy the aliasing :)

rp, maybe a stupid idea but why not put a (precise) bitcrusher on the master track and work with that for previewing?

rp, regarding "what does really get lost" : in theory: only dynamic range. think "6db per bit". if you apply dithering, you can get more than 8bit dynamic range out of an 8bit sample at the cost of a persistent noise floor. see montys videos at xiph.org, he explains it well. in practice, when using pro tracker, you'll lose more due to mixing and you'll get fancy aliasing. as virgill suggested, it's best to understand that as a feature ;)

@jco: I see the problem with mixing, but with no more than 4 channels? Also what do you think if only one instrument used dithering, especially a drum/break loop (which seems used on some great Amiga demo). I don't even know if I could hear a difference but.. ;)

(forgot to say it's possible to have different dithering on each instruments, maybe it create less problem, maybe more :P)

Are you going to use samples the classic way as instruments (transposed with every note) or as prerendered longer loops?
Because your main concern is sample rate, not bit depth.

Bit depth conversion from 16 to 8 bits should be hardly distinguishable if done correctly. Did you try open source SOX? I think it even supports IFF format...

If you're going to use loops, resample them to the highest sample rate supported by ProTracker - then you could even use dithering noise shaped to frequencies where they will be out of human hearing.

If you're going the classic way, don't use dithering as it will add too much noise.
Still, sample rate is going to be your biggest enemy - you'll lose high frequencies. Fortunately Paula has such a crude way of playback that it adds nicely sounding aliasing artifacts when playing back at low sample rates...

The coder is your biggest enemy, because he will devour all the beautiful chip RAM that you had dreamed of putting your hifi-loops in.

Make your coder stream samples from FastRAM :P

bucketheas, dithering only works if it's the last step you're doing coming from higher depth material. using pre-dithered samples in protracker will be of no use as soon as you play anything else along with it, gain it, pitch it, or do whatever with it.

so, forget about dithering with regards to preprocessing samples

It's not like dither depends on being in bit 0 only (there's nothing magical about it), so there's nothing inherently wrong with gaining a dithered sample. But dither adds noise, and many operations will amplify noise in weird ways. In particular, noise shaping will no longer be shaped the way you want it after pitching.

The alternative to dithering (rounding) is nonlinear distortion, which isn't much better.

sesse, "adding noise" comes with the advantage that you gain a bit of dynamic range due to details being spread out in a temporal way and due to the way da conversion works. just mentioning it because it took me years to understand why rising the noisefloor can actually be a good thing ;) nevertheless, it doesn't help with the topic this thread is about: making samples sound cool when used in pro tracker, with all the protrackeryness.

also, dithering is not "rounding" ;) it's adding noise with a bias.

anyways, due to the low dynamic range 8 bit provides, I'd try to make the best use possible of those bits, by heavily compressing the samples to maintain as much detail as possible BEFORE mixing in protracker or any other 8bit mod player/tracker.

sesse, sry i might have understood you incorrectly: if by rounding you mean "adding noise at higher bit depth" and afterwards rounding, then you are correct of course. it leads to the noise being shaped in a way that sub-bit dynamix range can be achieved. and indeed it's a rather simple process.

haven't heard about nonlinear distortion being able to achieve similar effects though, would love some urls about that topic. it does give me some ideas though, probably comes with the penalty of freq dependent aliasing?

My first though was someone should get Amiga output IR and then I remembered Plogue Chipcrusher. Has anyone done comparison between that and actual Amiga output? Would be a good way to monitor what samples will actually sound like, if it sounds authentic enough.

i googled and found linky. kb, the article from your blog this forum links to, do you still have it? seems to be gone...

Thank you, great replies here!

They helped to see my issue more clearly: lack of control and workflow. For the current project, I tweaked the main drum loops' dynamics again and again in the DAW (16bit). And the conversion to 8bit changed the overall feel a lot. So I'm stuck in a loop of tweaking, exporting and converting.

It'd be much easier to nail it in an 8bit environment.

@jco - Had the bitcrushing idea already as well and tried that. But how much sense does that really make - since the output from the DAW is 16bit dynamic range? Smells like meta-trouble to me.

A PAULA VST which properly emulates the Amigas sound with its different sampling rates and direct sample export would be awesome - wouldn't it?

I mean, the trouble I've got is probably related to Nyquist frequency and aliasing crap. Especially when trying to get the most bang out of looped sequences it all goes haywire.

And yes, I'm still stuck in a 90s mindset, that modules should be as small as possible :) I think that music on the Amiga could sound better and more balanced if one would be able to prepare samples with more knowing than guessing.

Maybe I'm just a lazy fuck.

You could use D16's Decimort 2. Set it to 8bit, ~60% dithering, resampler freq to 22050Hz (definitely not more than ~28kHz).
Route the output through a hi-cut eq that's set to ~4.9kHz / 6dB slope (A500 LED off) or ~3.2kHz / 12dB slope (A500 LED on).

This gives you a decent approximation of the Amiga sound (at least it's much easier than tweaking+exporting over and over again).

Like other have mentioned you definitely want to reduce the dynamic range of your 8bit samples to ~48dB, so use a compressor.

I just spent an hour fiddling with this, and yes, it's quite hard. It also brings back some not-so-fond mod tracking memories ;)

What's especially problematic are the low frequencies (e.g. bass drums).

You could use a short transient sample followed by a single cycle loop (triangle/sine, for the release phase), coupled with a volume+pitch envelope (i.e. do this in realtime on the Amiga).
If your tracker does not have a combined volume+pitch slide effect (or even an envelope), just use the A0x effect (in protracker), although it sounds much nicer with a pitch slide.
This should help you get rid of the horrible noise during a kick's release phase.

totally forgot about this (!)

bsp, nice suggestions there. tbh I have failed at "good conversion for protracker" samples myself... the stuff I exported tended to sound dull and lacked edge. so rp if you find a good streamlined workflow, please let us know.

also, a magic vst that could be hooked up in a daw and has a small "capture, optimize and export for kickass 8bit mod sound" button would be cool... hmmm!

Let's chatter about it at UC.?!

sure thing :D