ashcat_lt said:
It's dynamic range, which is the only real meaningful definition of the term "audio resolution".
Gotta respectfully disagree. In digital audio, resolution is largely based on the bit depth. To a small extent it can also be based on sample rates, but that has mostly to do with Nyquist level frequency limits and the "stretch" if you will, of the limits of a reconstruction filter slope. Which is not even really the point. The simple observation by many people is that plugins often sound better at certain sample rates. The big tradeoff with sample rates is drive space and overall processing power.
Bit depth is easier on both, and often used as a synonym of resolution with PCM audio.
In theory, 16 bit fixed point provides 96 dB range. This is better than an Ampex or Studer. With a super clean system and good dithering practice the range can be extended to around 108 dB. This would be equivalent to the theoretical range of 18 bit. At 20 bits you have 120 dB range. And a physical barrier at the thermal limit. Nothing we can make can surpass this range.
And yet we have converters that can record in typically 16 or 24 bit. Why would we need the resolution of 24 bits when the range is outside the thermal limit? There is more going on than dynamic range, perhaps.
Add to that the ADC/DAC processes typically ONLY operate at 16 or 24 fixed bit depths. This is different than processing.
I can get a 16 bit mix going with a bunch of compression, some reverb, maybe some EQ or M/S processing or whatever, and throw a dither plugin on the 2 buss. We know that quantization errors in 16 bit (or any bit depth for that matter) cause distortion, and dither kills the distortion but adds noise. But when I toggle the dither on and off, I can't hear any noise or distortion at either setting. What I can hear is the snare. With dither on, it sounds full and I can hear the whole reverb tail. It has its own distinct place in the mix. As does everything else. When I bypass the dither, half the decay of the snare is gone. The stereo image collapses. Things sound less distinct and more like cardboard. This is the onset of harsh digital crap from artifacts and the only difference is a plugin at the end of the chain that you barely have to pay attention to.
It's been described as laziness to avoid it. The Pro Tools manual I have doesn't help either. It says don't dither 24 bit tracks. What it doesn't say is that you might not be able to notice for a while, but the artifacts left behind could stack up after multiple processes and come back to bite you in the ass.
At the extreme end of the scale you have MP3's that have been invaded by the swarming space goblins. Same problem, left unchecked through multiple stages. The sonic decimation remix.
A DAW will run its internal processes at higher resolution to both the source and target formats for processing. Typically 32 bit float, but could also be 48 bit fixed, like Pro Tools HD systems before version 10, or 64 bit double precision as is becoming more popular now. This helps a lot with applying gain changes. If all you do in a DAW mix engine is change a level, you're forcing the audio to be requantized. 32 bit float doesn't matter so much, it's basically the same as 24 fixed, but with 8 bits that can scale the resolution up or down as needed for volume changes. If the change in volume is precisely 1 bit or 6 dB, dither isn't necessary because the samples will line up in the same spot. No quantization error. Anything else will cause problems unless your DAC can work with 32 bit float.
Add to that any plugins you might be running will probably not be using a floating point system for processing as it becomes difficult to impliment. So they might change the incoming signal to 64 bit fixed on the fly. When it's done, dither should be applied when it goes back to the mix engine. It just happens (or not), they don't tell you. It's just that some plugins sound better than others.
If I want I can quantize Pi to an interger value of 3. Couldn't possibly affect the circumference of a circle, could it?
Data resolution from bit depth is real, and it has a real effect on the processing capability of your audio. It's EASY to have errors in the math downstream and while in any practical sense the dynamic range is still governed by the thermal limits, there is just more going on than dynamic range. I've never heard of a 20 bit fixed point mix engine, but that's the practical limit of dynamic range we have to deal with.
There's a reason.