Why can't I distinguish bit depth beyond 16-bit? I have excellent hearing (even for a 52-yo)

I was reading a thread in which the topic discussed was 32-bit floating point bit depth. It got me thinking about how I can't hear the difference even beyond 16-bit. I've never read anything regarding that before.

I bounce/render my music out at 24 bits because it's the standard, but not because, "Oh this CD sound is a little noisy compared to 24-bit". My understanding is that the noise floor is lowered significantly with higher bit rates, I just can't hear it.

This is a hobby to me now, but how screwed am I?

 

Zero. I can't either.
Nobody can unless you use a wall of nuclear-atomic-speakers and crank up the volume cowboy-style on a very quiet passage. Then again, you will end deaf if you hear any non-quiet passage at that volume...

Ok. I overdid it a little bit but that's the gist of it

 

Last edited: May 29, 2024

You record at 24-bits because it gives you headroom, not because it sounds "better". If you record at 16-bits, you now have to struggle to keep the signal high up, without clipping, because recording at lower levels really does reduce the audible quality (you're essentially losing an effective bit of resolution for every 6dB lower than 0dBFS you record at). This is what we had to do back in the eg DAT days.

24-bits gives you ample dynamic range than you don't have to try to track up at 0dBFS and risk clipping, and still retain as much signal fidelity as possible.

 

24bits dynamic range is SO high it is mostly used for tracking or almost uncompressed music (like old school symphonic music).
That's why 16bits still the most used release format.

As a side note, Thriller was tracked in 16 bits :D

 

I can agree with that as well. Bit depth for me seems all relative. However I can hear a difference in the "hardness" quality of audio at higher sample rates. Like 48k vs 88k vs 96k etc... And my hearing is not any better than anyone else's here - maybe worse. It's more of a psycho-acoustic thing than anything else. I was just doing a kind of comparison the other day with some different sample rate stuff on Qobuz - and for most genre's comparing a 44khz sample rate master to a 96Khz sample rate master, the "hardness quality" (I don't know how else to describe it) of the audio is much more evident at 96khz. Of course the 96khz masters were @24bit as well so who knows... The playback hardware was of no consequence as I was listening to the stuff on a logitech desktop system. Going beyond 96khz for me is all relative once again.

 

Last edited: May 27, 2024

Bit depth over 16-bit gives more information UNDER -96 dB Fs. So if you want to hear that, set up a 10 kW speaker system in your living room and crank it up

 

you may take a track at 24bit 44.1khz, convert it to 16 bit, 44.1khz, and it may not be that easy to tell the difference. But if the same song was produced in a 16bit 44.1khz DAW session vs. a 24bit 44.1khz session; you would absolutely hear the difference in the outcome. This is like the silly "my free DAW can play a file back the same as your expensive DAW" topic that people often discuss ad infinitum.

 

Short answer: because you can't.
Long answer: because it makes no sense to use all that headroom. Even in the most dynamic music in the world, it doesn't make any sense.

Kinda crazy how I managed to link this video twice in less than a month!

 

humans can't 'hear' beyond 16-bit depth. the hight depths are normally used to have more headroom in the creation process, since it ''destroys'' the audio without much damage in the final mix/master (thats why, as they mentioned, you cannot hear the difference from 24 to 16... but you will hear the difference of the opposite, and this is because there has to be created data to fill in the empty data spaces on the audio

 

Hardly anyone can hear the dynamic difference of 16bit (96dB DR) and 24bit (144dB DR), because the perception of dynamic range of humans is around 140dB (from a pin dropped in the room to a rocket being launched).
Besides, your audio interface is highly likely using a 24bit converter.

 

Last edited: May 28, 2024

1. What I can hear with my audio interfaces is that I hear difference of 18 vs 19 bits with maximum volume on quietest tails etc.
2. All that stuff is headroom of not letting digital noise and quantisation break thru into being perceived levels. Headroom from under.
3. 32 bit float as I think is 24 bit with saving any digital levels above 0 dBFS.
4. There is also 64 bit float, for even more less noise and more headroom above 0 dBFS.
5. Dithering and noise shaping also give us less distortion, less noise (but shaped), no quantisation shit, more levels.
6. Who cares? Your levels will be lowered by normalization to -14 LUFS or so. Dithering and lossy codecs will not save from quantisation error. Your songs will be lowered to be quieter and quantized with you your dithering and "just only noise". Dithering only for lossless only.
7. Mp3 (don't know about others) is a float codec, but not guaranteed. Aimp reads as non-noisy. Reaper and RX will show and monitor as noisy. Everything is different.
8.Do music and sounds to be best perceived and HQ.
Many questions still.
But don't overclip, overlimit, oversquash your masters. Glue is good, but streaming, digital distribution will do the shit. Everything is against loudness war.
Imho, don't go for -5 LUFS integrated, keep that all between or at -8 up to -11.
All processing in ITB is, by the way, minimum 32 bit float, maximum 64 bit float, sometimes more (marketing 80 bit float, 117 etc...). Keep to be good at 90-105 dB or so to be clean.
buy AD for best recording to be less noisy, more dynamic range etc, best for headroom and noise and further processing.
I wish that all that be at 18-20 bit final etc and 480 kbps mp3 lossy. digital is digital. Find compromises and best quality, levels. Don't overdo. Don't be next Skrillex.

 

Last edited: May 27, 2024

Because you're thinking about it the wrong way around.

There is no difference between 16, 24, 32 and 64 bit sampling as long as the level of the signal stays above -96 dBfs.
The number of bits determines the minimum loudness of the quietest possible sound that is not silence.

This is why you can sample at 12 bit or even less and there will be no difference until the signal level drops below the minimum loudness of 12 bits (which is about -72 dBfs).

You can determine how loud the quietest possible sound will be by multiplying the bits by 6.02.

6.02*12 ~= 72
6.02*16 ~= 96
6.02*24 ~= 144
6.02*32 ~= 192

Try putting a bitcrusher on a track and observe how levels shoot up as you drop to really low bits. Until you get to 1 bit and there's only clipping or absolute silence left.

(Note that * 6.02 only works for integer formats, floating point formats have a larger range)

 

Last edited: May 27, 2024

You need a good pair of headphones,
something Detailed and Clear like DT880 or something.

With mine I can absolutely tell the difference.

Without them I'd have to have great speakers and listen very effing closely, much much harder to catch. It's very Subtle.

At that level with the speakers resonating in a room, it's more about the dynamics and feel than sound fidelity per se..

 

Last edited: May 27, 2024

What surely doesn't give you more fidelity is 32bit audio,
at least the way it's usually implemented..

A 32 bit file is made of 23 bits of mantissa, 1 bit of parity, and 8 bits of exponent.

The mantissa is the audio data per se,
what the exponent gives you is much more dynamic range.. so much so, it's virtually infinite.

That's why 32bit handheld recorders don't have a volume control for recording,
they record everything.. or at least everything the microphone is capable of detecting.

 

Lmao, I mix with $300 headphones plugged into a $150 headphone amp that goes into an Yamaha mixer. There's no difference because no one leaves that much dynamic range in. Maybe if you're listening to crickets in the forest, but even that probably has less dynamic range.

 

I know that i dont know. Period.

Some gear manufacturers had battles on gs forum prior to industry started blowing up and it get´s spooky with the audio thing overall, how much one can limit himself just by thinking while listening is incredible. I mean, we are basically ghosts and we percieve, we do not "hear", we percieve it thru our brain...so we have a "buss" thru which everything get´s in and that freeakin thing changes from vitamin D or C levels, from mood, from sleep/lack, from all kinds of things.

On a good day you could maybe hear it, on top end system let˙s say, or with recordings that were tracked with the full range like from old neve consoles that have above 50k freq response and so on, and probably the kick would have a bigger thump, or vocal would poke out more.

It is the same thing with the daw-s in my opinion, people who compose like Zimmer and Junkie XL say Cubase sounds the best, Timbaland says Ableton from 10-11 sounds good and you cant say they dont know what they are saying. So to get back to coding and all this stuff, i´m certain that ´Arguru´ was an advanced coder, and i tried his trackers like Psycle and MadTracker2 and they sound really good, but i´m not going to argue with nobody about daw-s, but i think i would want a coder who has really good ears to make stuff. I think he collaborated on FL Studio sound engine at some point.

Now Ethan really took time to think and write, i think he has some good insights
https://ethanwiner.com/perception.htm

This is with the link for his audio tests...that went thru youtube compression....i mean, top notch thinker can also have blind spots.
https://ethanwiner.com/audibility.html

 

Last edited: May 28, 2024

It's summing engines and quantisation. As more channels are summed, the output level increases—a higher bit depth permits more accumulated volume and permits for better application of plugin effects without introducing quantisation so early.

Apply reverb on a 16-bit engine and you'll hear the quantisation noise on the RT60.

 

Use your ears, bro.

 

This is true, but to find an audio engine that runs on 16bit integers it's pretty hard these days... Something like Fasttracker II on a 486 w/ a Soundblaster 16 would do I think.

No plugin standard offers this and even 32 bit floats can represent 16 and 24 bit waves exactly, though most modern engines go with 64 bit double floats anyway and plugins go with those.

I highly doubt there's any human that could distinguish between float->double conversion...

 

because ambience noise floor in your room is too high,
frankly most people won't hear a difference between 16bit and 8bit because -45dBfs noise floor is above minimum level of casual listening dynamic range people are able to dial in on nearfield monitors