Why is 16 to 24 bit no big deal - no popups for conversion?

Audio Plugin Hosts and other audio software applications discussion
RELATED
PRODUCTS

Post

Burillo wrote:
lfm wrote:I saw some expert panels from Pensados Place at some exhibition fair, or whatever it's called. The renowned awarded mixers and producers told how they moved everything to tape first.

Digital synths was a blessing in the sense oscillators didn't wander off - and needed constant retuning the first couple of hours it was running.

And we all know the amount of analog synths that are back with the analog things

It's not like - in the beginning there were nothing but ones and zeros ;)

Any more input, please just throw it here....
all of that has nothing whatsoever to do with the topic at hand.
It's my topic - so I decide, OK!

Since it was too far fetched for some that belive the 16-bit digital is so perfect.
You loose nothing, and there is nothing to rescue.
Continue belive that, it's ok.

On the ongoing digital revolution there are plenty more things to arise.
New conversions to higher audio bitdepth with closer resemblance to original may be one them.

Various audio restore procedures exist already.

It's not like upscaling from DVD to 1080, or 1080 to 4k is unheard of.
Both these examples are better looking than on original resolution screens - done right.

There are smart algos for upscaling photos to printer resolution also.
Also downsizing photos 6000x4000 to video for time lapse video - using smart algos.
That is what I use for photo and film.

So why not smart upscaling bitdepth for audio????
Say just getting every sample probably 20% closer to original is a big win.

I'm pretty sure it exist, just not on the price levels we usually handle audio in daws. Restoration stuff of all sorts, old archives.

In real life it's not a biggy for me, I re-record most stuff. Just curious...

Post

whyterabbyt wrote:
sonicpowa wrote:
lfm wrote:Is there maybe a way to make better conversion going up in bit depth. I think there might.
Are you thinking there is a way to somehow restore those once truncated/dithered bits? This is of course impossible.
Meh; of course its possible. The only reason we dont just store everything as 1-bit/44K and interpolate back up to 24-bit is because of the hard-drive manufacturer's propaganda.
I don't usually speculate on audio voodoo, but putting my tin-foil hat on for a moment, I wonder if this might actually be possible.

Good dithering lets you represent signals well below -96db in a 16 bit audio file, so it's doesn't seem beyond the realms of possibility that those signals could be reconstructed if there's an algorithm that can distinguish between dither noise and signal.

As a thought experiment, maybe if we were working at 96kHz 16 bit - we filter off everything above 22kHz, then push all our dithering noise above 22kHz - we can call everything below 22kHz "signal", and everything above 22kHz "dither"... would that make is easier for our pointless imaginary algorithm to meaning fully reconstruct some signal if we convert to 44kHz 24 bit?

I mean, I don't think it'd be particularly useful, but just thinking out loud here.
Last edited by cron on Wed Jul 26, 2017 5:50 pm, edited 1 time in total.

Post

lfm wrote:Since it was too far fetched for some that belive the 16-bit digital is so perfect.
You loose nothing, and there is nothing to rescue.
Continue belive that, it's ok.

On the ongoing digital revolution there are plenty more things to arise.
New conversions to higher audio bitdepth with closer resemblance to original may be one them.
...
In real life it's not a biggy for me, I re-record most stuff. Just curious...
I've explained this in my above post. The 'grid' above -96dB is exactly the same in a 16 bit file as a 24 bit file. You can't upsample it because you're not adding any new points above -96db when you convert to 24 bit.

There are PCM formats which push the quantisation 'points' higher up the scale, so there are more at the top than there are at the bottom (e.g mu-law, A-law) but these were only really used in telephony and ultra-low bit-depths where it makes more sense to have higher resolution for louder signals than quiet signals. But the linear PCM we use today where the quantisation 'grid' is evenly spaced makes far more sense from a high resolution audio perspective.
Last edited by cron on Wed Jul 26, 2017 5:56 pm, edited 3 times in total.

Post

The problem you get is that the brain is extremely sensitive to artificially generated artefacts in audio. No-one believed the MP3 birdies were going to be a problem until listeners in tests complained about them. There's a story recounted by Paul Frindle at Gearslutz about dither problems they encountered at Sony. OK that's going the other way, but on paper the dither algorithm looked good. It just sounded unnatural to a small but important group of testers.

The brain is far less sensitive to high-frequency artefacts in images. This is why upscaling works (up to a point) in the visual domain. The brain looks for edges and pretty much smoths everything else out. You can hide all kinds of crap in images that, were you to do the same in audio, would be immediately apparent to many people.

And you'd risk introducing artefacts to gain, at best, a few bits of 'lost' detail. And you'd probably have a whole new slew of "Paul is dead" or "worship Satan" myths appearing about famous recordings after they'd been 'upscaled'.

Post

sonicpowa wrote:
lfm wrote:Is there maybe a way to make better conversion going up in bit depth. I think there might.
Are you thinking there is a way to somehow restore those once truncated/dithered bits? This is of course impossible.
Not on bit level - let's say closer to original than 256 step bumps!!!!

If analyzing material properly you might find 5-6 bits accuracy.
Instead of 256 steps, you get as close to 32 or 64 steps where you try to decide what is better. I think that already matters.

The smart algos for dithering do a lot of analyzing what gives better probability for that music material. When dithering was young one size fits all, kind of. And it worked worse for some material than others. it's evolving all the time this knowhow.

And thinking the other way too, might there be something to gain.

When people have their enormous archives of CD's they don't bother listen to anymore because they sound so horrible - maybe there will be a service that restore to better sounding ones.

You send in the bunch, and get back a 4TB thumb drive all in SACD resolution.

One way today is probably to drown everything in harmonics adapted to hearing.
That fool some, I'm sure.

Dont' know status of Neil Young's Pono project. It was crowd founded I think.

Convenience goes a long way, and mp3 is good enough for most listening devices.
But would like to listen to Pono if I had the chance.

Post

lfm wrote:Since it was too far fetched for some that belive the 16-bit digital is so perfect.
You loose nothing, and there is nothing to rescue.
Continue belive that, it's ok.
why do you insist on this stupid straw man argument? what logic do you use to get from "nothing is lost when upsampling from 16 bit to 24 bit" to "16 bit is perfect"?
lfm wrote:On the ongoing digital revolution there are plenty more things to arise.
New conversions to higher audio bitdepth with closer resemblance to original may be one them.
since you appear to not know first thing about digital audio, i wonder where do you get that confidence from.
lfm wrote:It's not like upscaling from DVD to 1080, or 1080 to 4k is unheard of.

Both these examples are better looking than on original resolution screens - done right.

There are smart algos for upscaling photos to printer resolution also.
Also downsizing photos 6000x4000 to video for time lapse video - using smart algos.
That is what I use for photo and film.

So why not smart upscaling bitdepth for audio????
so you basically ignored everything i said back in my first post, about how increasing bit depth is not at all like upscaling and is in no way analogous to it. but yeah, go ahead and regurgitate this nonsensical talking point.

if, however, you want to actually learn something from this discussion, say it with me: increasing bit depth is not at all like upscaling. increasing bit depth is not at all like upscaling. once again, with feeling: increasing bit depth is not at all like upscaling.

you say you work with photo and film - great, now tell me what happens if you convert an image from 24-bit color to 32-bit color. and while you're at it, you can also tell me that because you can losslessly convert 24-bit image into a 32-bit one, 24-bit images are therefore perfect - you love to make this nonsensical point about audio, so you might as well do so for images.
I don't know what to write here that won't be censored, as I can only speak in profanity.

Post

Gamma-UT wrote:The problem you get is that the brain is extremely sensitive to artificially generated artefacts in audio. No-one believed the MP3 birdies were going to be a problem until listeners in tests complained about them. There's a story recounted by Paul Frindle at Gearslutz about dither problems they encountered at Sony. OK that's going the other way, but on paper the dither algorithm looked good. It just sounded unnatural to a small but important group of testers.

The brain is far less sensitive to high-frequency artefacts in images. This is why upscaling works (up to a point) in the visual domain. The brain looks for edges and pretty much smoths everything else out. You can hide all kinds of crap in images that, were you to do the same in audio, would be immediately apparent to many people.

And you'd risk introducing artefacts to gain, at best, a few bits of 'lost' detail. And you'd probably have a whole new slew of "Paul is dead" or "worship Satan" myths appearing about famous recordings after they'd been 'upscaled'.
You may be right all the way.

An image is creating objects we know about, and what is stored about those objects before become some kind of memory that is what we see. It's all interaction between objects that we look for.

I had a media player, with I think 1" display, not much more - and playing video worked much better than I anticipated.

But what you say is also said about all amp sims, and tube and tape emulators. But it is getting better all the time.

Even though I am suprised turntables of today are sold with wow/flutter specs that nobody would even buy back in the day, before CD's. So marketing goes a long way, I suppose.

So restore procedures might go the same way as tube and amp sims.
- Cool, but that is not sounding right
at first, but improved over time.

I modified a DAC I have according to a Lampizator project, and this guy build high end DAC's today with tube amps. It brings something that our ears+brains like better than the all clean stuff. Cover bad stuff with more good stuff.

I guess that can be applied to imported 16-bit stuff too. Cover over it with a good tape emulator.

Post

Burillo wrote: so you basically ignored everything i said back in my first post, about how increasing bit depth is not at all like upscaling and is in no way analogous to it. but yeah, go ahead and regurgitate this nonsensical talking point.
So you wonder why nobody cares what you write.

I wonder why that is? ;)

Post

Digital is only as good as the cables.

Go ahead, I'm already swiss-cheese, what's a few more holes? :hihi:

Post

lfm wrote:So you wonder why nobody cares what you write.

I wonder why that is? ;)
i think it's demonstrable that in this thread, you pretty much don't care what anyone writes but yourself, and anything that contradicts whatever it is that you already imagined will go in one ear and out from the other. i won't try at guessing why that is.
I don't know what to write here that won't be censored, as I can only speak in profanity.

Post

The answer to the question has already been given many times.

Again: The quantisation 'grid' above -96dB is exactly the same in a 16 bit file as a 24 bit file. You can't 'upsample' it because you're not adding any new points above -96db when you convert from 16 to 24 bit. All the new points you add by moving to 24 bit sit below -96dB. Any hypothetical upsampling would only improve the parts of the signal sitting below -96dB. The stuff above that can not be improved in any way because there's no extra resolution in 24 bit above -96dB compared to 16 bit. That goes for recording directly to 24 bit as well as converting from 16 to 24.
Last edited by cron on Wed Jul 26, 2017 8:01 pm, edited 1 time in total.

Post

Just do the damn phase reverse check, there's your proof.
Put a 16-bit and 24-bit file side by side, flip the phase on one, hit play.
No measurable (implying: definitely no audible) difference between a 16-bit file that was "extended" to 24-bit depth.

All 24-bit is doing is adding more post-comma decimal floating points for higher precision calculations.
The value of 0.001 is always 0.001, no matter if you write it as 0.001 or 0.00100000. It's always 0.001.
The result of 0.01 * 0.1 is always 0.001, no matter if you write it as 0.01 * 0.1 or as 0.01000 * 0.10000. It's always 0.001.
Just like the additional unused 00000 are not relevant to mathematics, they're not relevant to the sound.

Until you start processing the samples.

If a sample's value (=amplitude, volume) can be reduced from, say, 0.001 to 0.000000001 by an EQ or a compressor or a closing gate, because that level of precision can be used to calculate, then that is definitely a different result than if, say, 0.000001 was the maximum level of precision your compressor or gate could calculate, and thereby it would be a measurable and (potentially) audible difference.

If you export the higher-precision result to a lower precision format, like maybe a 24-bit project to a 16-bit .wav file, then the lower precision will truncate the higher precision result. Using above numbers, if the high-precision result is 0.000000001 but you export to a format that can only represent 0.000001 precision, then the resulting lower-precision export would contain a truncated or rounded value.

Which is where dithering comes into play, but that's an entirely different minefield not worth walking into in this thread, where the OP clearly has yet to understand the differences between bit depths themselves.

Say your interface and its converters could actually record total silence, without adding "circuit hiss" or distortion from A/D conversion and the involved LP filtering and all that. Say you record several takes of total silence at 16-bit signal precision and leave the recordings otherwise unprocessed.

If you mix/stack those several recordings of unprocessed 16-bit silence, they each have a maximum precision of -96 dBfs, and all stacked on top of each other, their -96 dBfs noise floors will add up to a really audible rug of fuzz in your final result pretty quickly.

If you put a gate plugin onto each of those otherwise unprocessed 16-bit silence tracks, and you use that gate to totally mute every 16-bit track down to complete silence - then (mindf*ck) the gate's output will still not be able to reach anything lower than -96 dBfs, because that's the maximum precision of 16-bit audio.

So in effect, it will change nothing. The -96 dBfs noise floor remains. You stack a bunch of -96 dBfs noise floors on top of each other, you'll hear it fairly quickly.

Now, say all of the above scenario is true for this next example, we're dealing with the same 16-bit silence recordings and their -96 dBfs noise floors. The only difference is that the project is calculated at 24-bit precision.

Some maths: 24-bit precision means that the quietest value that can be calculated and stored is -144 dBfs. That's a difference of 48 dBfs. Depending on how you calculate it, using 3 dB or 6 dB as "half volume", then that would mean silence at 24-bit depth will be PERCEIVED at 48/6=8 or 48/3=16 times QUIETER than 16-bit depth.

What that means is: if you throw a bunch of 16-bit silence recordings with -96 dBfs on top of each other and calculate their mix/sum at 24-bits, nothing -absolutely nothing: not audibly, not measurably- changes. The quietest value will still be -96 dBfs, because the 16-bit source files don't contain any information beneath -96 dBfs.

However, if you now throw the same gate plugins as before onto all those 16-bit / -96 dBfs source tracks, and all the gates are closed, then their closed output volume (=silence) will NOT be -96 dBfs as before, but it will be -144 dBfs. Because that's the quietest volume, the highest precision a 24-bit project can calculate and represent, that's the lowest sample "volume" and the highest signal precision a 24-bit audio file can store.

What that means is: you're no longer stacking -96 dBfs gate outputs on top of each other as with 16-bit. You're now stacking -144 dBfs gate outputs on top of each other, because that's 24-bit precision.

TL;DR Conclusion: source files recorded at low precision 16-bit will NOT be audibly or even measurably altered when played back or mixed at 24-bit precision. But if they are processed by 24-bit plugins or algorithms (pan, etc.), then what comes out at the end WILL definitely sound different against what would have been the result with 16-bit processing.

One quick add: this depends on the host software, but usually if you load a 16-bit sample into a 24-bit project, it will not be converted or re-encoded or anything at all. It will remain exactly as it is on disk.

/thread
Last edited by Rockatansky on Wed Jul 26, 2017 8:06 pm, edited 1 time in total.
Confucamus.

Post

CD sounding horrible? :nutter: This is a trolling thread for sure...
Blind listening tests have proven why SACD was a failure.
We are the KVR collective. Resistance is futile. You will be assimilated. Image
My MusicCalc is served over https!!

Post

BertKoor wrote:CD sounding horrible? :nutter: This is a trolling thread for sure...
Blind listening tests have proven why SACD was a failure.
I dunno, according to Neil Young, they already lost everything, of course you could duplicate it forever.

:hihi:

Post

I'm far from a digital scientist and try not to get caught up in this kind of minutia, by my understanding, which I'm sure will be corrected if it's wrong, is that if you convert a 16-bit file to 24-bit it just adds zeros. Same thing as what happens to a unity 24-bit file file in a 32-bit float host, it will just add zeros to the stream, nothing changes, until some DSP processing happens.

[/not a digital science dude]

Post Reply

Return to “Hosts & Applications (Sequencers, DAWs, Audio Editors, etc.)”