What's the possible purpose of these large audio formats?

Question

I am learning more about how much data is actually necessary for high fidelity audio. As far as I can see, the highest quality data files able to be worked with and exported from Adobe Audition (CS6) are WAV-PCM and BWF-PCM files as described below:
Format: Wave PCM (*.wav, *.bwf)
Sample Type: 192000 Hz, 32-bit
Bitrate: 24576 kbps
Format Settings: Wave Uncompressed, 64-bit Floating Point (IEEE)
Markers and other metadata included
Large estimated file sizes, such as 6 min = 1 GB for stereo audio
Multichannel audio available
44.1kHz at 192kbps (generally 16-bit) is usually the maximum level of audible fidelity. I prefer a minimum of 256kbps and 24-bit (32-bit if possible). What's the purpose of such large and complex audio file formats?

Answer 1

steve.kr wrote:
44.1kHz at 192kbps (generally 16-bit) is usually the maximum level of audible fidelity. I prefer a minimum of 256kbps and 24-bit (32-bit if possible). What's the purpose of such large and complex audio file formats?
Forget about the bit rates - they're irrelevant. You don't need to sample any faster than 44.1k because that rate will let you include all of the information present in the signal, up to about 22kHz. That's several kHz further than any adult hearing extends. Your ears have a dynamic range that is potentially greater than a 16-bit signal's, but not in one go. If you start at 0dB (using this as the hearing threshold), then at about +85dB the Acoustic Reflex mechanism kicks in, and progressively shuts down the sensitivity, but this allows you to hear sounds with higher levels without damage. But it desensitises you at lower levels when this happens, so the 96dB on offer from a linear 16-bit system is more than adequate - as has been extensively proven by research.
The point of greater bit depths has nothing to do with hearing, but everything to do with processing. If you use the 32-bit Floating Point system you can manipulate levels up and down, and even save the results inbetween, without loss. This won't work with an integer based system - you lose resolution at the point you save if you attenuate your signal. With the Floating Point system, you can whack it up and down without loss, because the changes are stored in the scaling bits, and the actual (24-bit) data is unchanged.
The history of all these extended formats is a bit murky, and based on a lot of misinformation spread by people who should have known better, but didn't. There was at one stage a technically sound reason why some higher sample rate recordings sounded 'better' - this being the dodgy brick-wall filters that were being used creating large phase errors in the audible frequency range. But now that oversampling is the norm, 44.1k systems work absolutely fine, as they should do, and there's absolutely no justification whatsoever for massively high sample rates unless you're recording bats. There is no meaningful musical harmonic content that extends beyond about 14kHz, and if you include a lot of harmonic content at higher levels, a lot of listeners find it quite disturbing.