Porting Miroslav Philharmonik is easy

[Otto Halmén]

(09-28-2017, 05:24 PM)Samulis Wrote: (the fancy schmancy duplicating and phase inverting type stuff I'm pretty sure wasn't possible in any sort of real time way on most of that old hardware, though many had haas-effect-style stereo 'creators', but those seemed to bake it into the samples if I recall right).

Those two are actually quite trivial to implement. Phase inversion is just inverting the sign in each signed PCM sample, and Haas panning or stereo upmixing is just adding a very short delay. Not sure which ones of the mid-90s samplers allowed to be configured for just that, but they were running complex stuff like digital reverbs (i.e. lots of simultaneus delays, filters, and LFOs) so it most certainly wasn't a matter of real time capability.

[Samulis]

Oh of course they were able to run lots of stuff real time... but some of them actually didn't have effects, such as the EMU ESI-4000 (which was like 1998 or so!), where you had to buy a daughter board to do any effects processing aside from baking in some DSP stuff. On the other hand, stuff like the Kurzweil, Yamaha, and even Technics of all people had extensive built-in effects that could be configured per instrument. I think that Technics modelling synth I have has more effects built in than I have effects plugins on my whole computer... though obviously there have been improvements since the late 90's in the quality and depth of DSP tools and effects.

Typically for most of these I've found that some sort of "stereo creation process" either involves baking it in or using some sort of built-in Haas type effect (some baked, others not). Many earlier ones back in the 80's just used chorus effects to create stereo, which actually sounds pretty decent on electronic instruments but uh... less than desirable on acoustic. By the mid 90's though, most companies were offering both mono and stereo versions of their instruments on a disk. The idea was that if you had the space, you would use the stereo samples, if you didn't, you got to use mono! D:

Which brings us back to today- there are actually a decent number of sample library developers who record in mono (yes, still!). They use very good convolution reverbs to "reinflate" the sample, sort of like the final stage in a modelling synth taking the generated audio and putting it in a space. It's a smart move because it cuts size in half and convo reverb can be very good at reinflating a signal- as long as you have a really good impulse to work from! A good example is ERA, which I am 90% sure is entirely mono in original recordings. I'd guess convolution was a bit too expensive for most libraries back in the pre- to early 2000's, so that's why they did this phase inverty/eqy stuff.

EDIT: I do have one question... what happens when you take the left and right audio output and overlap them with this phase inverty stuff? Does it cancel horribly?

[Otto Halmén]

Yes, a reverb would spatialize a mono instrument properly, especially if it's a small solo instrument. At one point, I tried running all orchestra tracks in mono just to see how much stereo samples really matter. They mattered slightly for the string section. For the rest, not so much. A reverb works wonders indeed.

Optimization is smart indeed. After all, hardly any virtual instrument gets used in isolation. There has to be room for a whole bunch of them.

It "worsens" the public image, though, which is unfortunate. If Miroslav Philharmonik would have been distributed in 16-bit instead of the utterly superfluous 32-bit, it would only have been a 3.5 GB library, not a 7GB one. Add in lossless compression, and you might drop below 2 GB. Small numbers just don't feel as good, I suppose.

As to my test, no, they don't cancel out completely, although the cancellation is still observable.

[Samulis]

I still think that 32-bit might be an Audition bug, but yeah... the whole public image thing is annoying to work around. I still get a bit miffed over people getting upset that some stuff in VSCO 2 is 16-bit and not 24-bit. I typically just do 24-bit for everything now just to avoid all of that.

[Otto Halmén]

I get that 16-bit samples have a noise floor that, once you sum enough 16-bit samples, might rise to an audible level. Still, you pretty much never render each sample at full amplitude in a project (unless you really love clipping), so I can't imagine the summed noise floors ever rising above the noise floor of the final bounce. Most platforms use either a lossy codec or 16-bit PCM, and broadcast (which uses 24-bit PCM) has the rule about -23 LUFS which often makes the whole thing have a rather low amplitude to begin with.

The only thing I'd imagine that would really benefit from high bit depth samples (i.e. a lower noise floor) is DI heavy guitar samples (due to massive non-linear processing from amp sims), although I haven't tried that in practice, so the extra bits might turn out to be redundant even there.

[Samulis]

Yeah, most noise floor issues either happen on extremely quiet samples that have to be amplified in the actual instrument (which are mainly the fault of poor engineering, but can also just be the result of an extraordinary dynamic instrument) or in locations with naturally high noise floors- i.e. concert halls! If you hear noise in samples, I can almost guarantee it isn't the bit depth that's causing it. Bit depth isn't going to do anything to noise that's already part of the signal! High self-noise mics, preamps, occasionally cables, and most importantly - the space - are what cause noise issues. A lot of very high end libraries are very noisy, typically because of the latter, but not always- some expensive mics have self noise orders of magnitude above, say a Rode NT1.

Pianos are the most susceptible, but a lot of percussion instruments come in a close second place.

[Otto Halmén]

Exactly! Also, denoising concert hall noise might not even end up as a subjective improvement. In my experience, denoising is a messy and imprecise business.

Moreover, aesthetic tricks with the inevitable dithering noise are a thing. Airwindows has a bunch of different plugins that do creative noise shaping (instead of the typical minimization of audibility). My favorite is the one that makes the noise floor sound like tape noise. It's theoretically more audible than your typical noise shaping, although I still can't really discern it at 16 bits even with headphones (unless I turn up the volume to unbearable levels and listen to the very end of a sound fading out or something).