[imrazor]

I'm intimately familiar with 16-bit audio since I first bought a Pro Audio Spectrum 16 knock-off way back when. And I've heard of 24-bit audio, which some dismiss as marketing hype. But I recently dusted off an ancient sound card, downloaded some Vista-era audio drivers, and voila I was in business.

I started looking at the audio options in the Windows Control Panel (Settings now) and noticed this:

Never seen that before. Since most of the audio I listen to is standard CD rips (16-bit, 44.1KHz, 320Kbps) I doubt I'll extract any extra listening pleasure from 32-bit audio. But I am curious about it, and whether anyone can tell the difference from 16-bit, 24-bit and 32-bit.

 

[ScornDefeat]

I'm intimately familiar with 16-bit audio since I first bought a Pro Audio Spectrum 16 knock-off way back when. And I've heard of 24-bit audio, which some dismiss as marketing hype. But I recently dusted off an ancient sound card, downloaded some Vista-era audio drivers, and voila I was in business.

I started looking at the audio options in the Windows Control Panel (Settings now) and noticed this:



Never seen that before. Since most of the audio I listen to is standard CD rips (16-bit, 44.1KHz, 320Kbps) I doubt I'll extract any extra listening pleasure from 32-bit audio. But I am curious about it, and whether anyone can tell the difference from 16-bit, 24-bit and 32-bit.

You would need a very dynamic track, an exceptionally quiet listening environment, and a resolving system with very minimal noise floor, to accurately tell the difference between 16-bit and 24-bit audio.

There are no human ears sensitive enough to differentiate 24-bit and 32-bit audio.

 

[Roseval]

If you are doing digital volume control (or any another kind of DSP) it is beneficial to have a bigger word length.
If you have a 16 bit word and reduce volume by 48 dB you have only 8 bit of information left over and you need to dither as well.
Do the same with a 24 bit word and you still have the full resolution

11111111111111111
0000000011111111 -48 dbFS and 8 bits left

111111111111111100000000
000000001111111111111111 -48 dbFS and all 16 bits still in place.

Mutatis mutandis 24 bit recordings and 32 bit data path.
So set the bit depth to the max supported by your DAC.

 

[gregorio]

[1] And I've heard of 24-bit audio, which some dismiss as marketing hype.
[2] Never seen that before. Since most of the audio I listen to is standard CD rips (16-bit, 44.1KHz, 320Kbps) I doubt I'll extract any extra listening pleasure from 32-bit audio. But I am curious about it, and whether anyone can tell the difference from 16-bit, 24-bit and 32-bit.

1. 24bit is not entirely marketing hype. It serves a useful purpose when recording, as it allows significantly more headroom but as headroom isn't necessary or beneficial for distribution, then as a consumer distribution format 24bit is just marketing hype.

2. There's really no such thing as 32bit audio, just 16, 24 or 32bit container formats for audio. A handful of the most dynamic commercial music recordings are 12bit audio but the vast majority are 10bit audio (or less) and whether you put that 12bit audio into a 16, 24 or 32bit container obviously makes no audible difference.

[1] You would need a very dynamic track, an exceptionally quiet listening environment, and a resolving system with very minimal noise floor, to accurately tell the difference between 16-bit and 24-bit audio.
[2] There are no human ears sensitive enough to differentiate 24-bit and 32-bit audio.

1. Even with an extremely dynamic commercial music track and the quietest of listening environments/most resolving system, at any reasonable listening level it would not be possible "to accurately tell the difference between 16-bit and 24-bit".

2. Regardless of human ear sensitivity, there's no audio reproduction system that could output anywhere near that difference in the first place.

If you are doing digital volume control (or any another kind of DSP) it is beneficial to have a bigger word length.

While your post is entirely true, it's also irrelevant. If you are doing digital volume control (or any other kind of DSP) it can be relevant that the DSP environment is 32bit but not that the audio file that you input into that environment is. For example, if the OP chose 16bit output and then applied DSP (which typically is 32bit or at least 24bit) then the result would be your second example ("all 16bits are still in place"). Only if the DSP were 16bit would your first example occur, although it would also occur if the output of the device were set 24 or 32bit.

G

 

[Roseval]

My assumption is that if you set Windows audio to 16 bit, the audio path is 16 bit, not to be mistaken for the DSP.
The way Windows handle this is (imho of course)
Convert to float
Apply DSP
Apply dither
Convert float to 16 bit integer
hence the loss of resolution.

 

[gregorio]

My assumption is that if you set Windows audio to 16 bit, the audio path is 16 bit, not to be mistaken for the DSP.
The way Windows handle this is (imho of course)
Convert to float
Apply DSP
Apply dither
Convert float to 16 bit integer
hence the loss of resolution.

That depends on where you do the DSP and, what DSP you apply. For example:
Output from windows to DAC at 16bit,
DAC converts to 24bit.
Apply DSP (EQ and/or digital volume control for instance).
DAC converts 24bit result to analogue.
No loss of resolution.

Even applying the DSP in Windows and then outputting 16bit will not audibly loose resolution, unless you do something really silly such as hugely reduce the digital volume and then apply considerable amplification later in the chain to compensate. In such a case, 24 (or 32bit float) would be advisable over 16bit but you could still run into noise/resolution issues. As the typical/decent DAC has an output SNR of 110dB - 120dB, loosing (as per your example) 48dB could effectively reduce the SNR to 62dB after the compensating amplification and lower still if we factor in the self noise of an amp operating at relatively high gain.

If we're going to effectively re-engineer a distribution master, then either we need a little audio engineering knowledge or at least, avoid doing anything too extreme and screwing something up (gain-staging in the given example).

G

 

[theaudiologist1]

32bit files DO exist. But they are in WAV or Wavpack formats. Flac only does up to 24 bits.

 

[gregorio]

[1] 32bit files DO exist. But they are in WAV or Wavpack formats. Flac only does up to 24 bits.

No one is disputing that 32bit file formats exist and incidentally, AIFF is another. What doesn't exist is 32bit audio, commercial music recordings are almost never more than about 12bits, no where near the 32 bits the file formats allow.

G

 

[theaudiologist1]

No one is disputing that 32bit file formats exist and incidentally, AIFF is another. What doesn't exist is 32bit audio, commercial music recordings are almost never more than about 12bits, no where near the 32 bits the file formats allow.

G

I agree having 32/192 files is very useless, but upsampling and upbitdepthing (is that a word?) to higher resolutions and sample rates does push the noise further away from the audible range, and increases the noise floor.

 

[gregorio]

I agree having 32/192 files is very useless, but upsampling and upbitdepthing (is that a word?) to higher resolutions and sample rates does push the noise further away from the audible range, and increases the noise floor.

Unfortunately it doesn’t! If you have say a 44/16 file, with a digital noise floor of say -96db and you increase it to say 192/24, you will still have a noise floor of -96db. This is simple to test if you have an audio editor.

If your initial recording were say 192/24 then the digital noise floor would be at about -144dB. However, that would NOT be the noise floor of the recording! Due to analogue noise and acoustic noise (the noise floor of the recording venue) the noise floor of the mixed and mastered recording would virtually never be lower than about -70dB. More bits or a higher sample rate does NOT reduce this analogue + acoustic noise floor!

G