This topic is a never-ending topic on sites such as
https://gearsz.com
You could debate it their for your the rest of your existence, or just make music instead.
Since they constantly get new members, and the site is already pretty old and successful, the topic never dies.
Personally, I like to maximize bit resolution so that when I process my audio with DSP there is no generation loss due to repeated DSP use and/or repeated gain changes. Since I freeze everything to audio right away and do a lot of audio manipulation, I am relieved that my main DAW programs are better than in the past in this regard. I used to read a lot of technical articles and manuals about this type of thing, and I even had a few college courses on this exact topic field during the late 1990s/early 2000s. I kept studying digital audio topics before, during, and after that time.
But getting back on topic,
I really like working at 32-bit IEEE floating point WAV. It's a nice guarantee that I'm not likely to lose any data. Also, it's the default datastream for VST effects and instruments. I use Cockos REAPER alot with the engine configured for 64-bit IEEE floating point and everything else (renders, freezing, glueing, recording) to 32-bit IEEE floating point WAV. There's even the Broadcast WAV and WAV64 possibility which is nice to have even though I don't need them. I don't even bother using mono files most of the time either, unless it's a kickdrum track that I want to be able to see at a glance needs to be mono.
But after a lot of listening tests and plenty of successfully received tunes and not much RAM, I convert my entire drum sample library to 16-bit files. Percussion is already not as detectable for fidelity unless it's tonal, and due to psychoacoustic masking, usually the tails are not going to be noticed much for the types of sounds that I use. I've gotten away with a few 32 kHz sample rate percussion sounds, but it is detectable (although some like it that way, less listening fatigue from high frequencies).
So I usually use AudioMove (32-bit Windows freeware) to convert my whole drum sample library collection to 48 kHz 16-bit WAV after I've done all the sweetening and editing from within a 32-bit IEEE float (engine) editor such as OcenAudio (LINUX & Windows). A lot of people just use the 24-bit recording defaults in REAPER, but that's kind of legacy from drums and vocals recording and it makes sense because audio interfaces can't currently do any better than 24-bit because it's the current physical electronics limitation. They physics don't currently allow for better than 24-bit audio of electrical signals as we currently use them. Light waves, I dunno, radio waves, I dunno, sonar waves, I dunno. But for pro audio, 24-bit is the limit. For 2-inch tape, 20-bit is the limit.
Whether PCI or USB or Firewire or other, 24-bit is where inputs and outputs max out in terms of fidelity.
But anyways, since I downgrade my drum samples to 16-bit, alot more of them fit into RAM. In fact, my entire drum sample library, which is rather large by my own personal standards, shrinks to a 7zip file of only a few MiB's. That's glorious for me--it makes re-using that archive and storing it really easy. Others have gigabytes or terabytes of drum samples, but that's overkill to me. Perhaps they are using loops though. I'm referring to "one-shots".
In terms of sample rate, I stick with 48 kHz. For a while I liked the idea of 88.2 or 96 kHz so that I could do fancy pitch shift experiments several octaves with zero quality loss, but i've since quit doing that. Actually, a lot of the time, 8-bit 11 kHz audio sounds phenomonally interesting due to all the aliasing if it's done right, very much like decent bit-crusher effects.
I started out many years ago using 8 kHz 8-bit mono sound files on a Mac SE, so I know what they sound like (phenomenal!)
Since FLAC, WAVpack (WV), OPUS, M4A/AAC, OGG, and MP3 can all each do 48 kHz, I like that rate. Also, r8brain can work with that. And it's the DVD standard, so it's pretty popular that way. And it's compatible with Vimeo and YouTube other sites and services. And it's compatible with camcorders too. It's better than CD (RedBook) quality, yet still highly compatible and the filesizes aren't humongous. There's a slight improvement for latency at a relatively small cost in terms of CPU. It guarantees the whole humanaly-possible audio hearing range (10 Hz - 24 kHz / 20 Hz - 20 kHz) and it won't bother cats and dogs and bats or fish. It's the closest thing we've got to 64 kHz which some say would be ideal. 88.2 might be better, but most hardware that's kinda old, can't do 88.2 kHz, but can still do 48 kHz. Some old hardware actually works better at 48 than 44.1 kHz also! So I'm sold on 48 kHz. Any larger would be using up too much RAM, CPU, convertion time, and drive space of which I don't have a lot.
Last but not least, I read a nice article in some computer magazines that modern-day motherboard manufacturers now regularly have 24-bit 48/96 kHz ADC's and DAC's by default. They might not be as good as pro converters, but they are guaranteed to be better than previous generation converters in terms of noise and some capabilities. Yet again, maybe not all would work out in the real world at 96 kHz, in terms of the strain, but 48 kHz is still easily high-fidelity without bottlenecking the systems.
Windows Vista,7,8,10, and WINE can all handle 48 kHz. Macs can handle 48 kHz too, if I remember correctly. Many antique laptops and desktops can nowadays still do 48 kHz since the ones older than that are now usually in the landfills or hopefully recycled.
I like CD's a lot, just bought some recently, but I can see at the record stores that they aren't as popular as they used to be.
On the downside, many hardware lossless audio players that also play MP3's can't do 24-bit FLAC's unless they are modified by the RockBox project. But they can still do 16-bit FLAC's. Usually they can do 48 kHz.
So we are in good hands these days. 96 kHz still might have it's officianados, but i'm not one of them.
Once I owned BPM, the drum machine software, and it was very tedious because the EXTREMELY BLOATED soundsets were ALL at 96 kHz. But it was ridiculous because many of those drum samples had hiss and AC hum in them in a bad way. To make matters worse, they were in a proprietary format, and the software itself had latency and CPU problems as well as iLok fussiness.
If they had simply put the sounds into WAV or at least AIFF format, then at least I could've still been using the sounds which I liked, instead of throwing the whole thing away. It was a mess. Too bad, because the software was great in other ways, but not much greater than FL Studio's step editor, in fact. Beat Thang kinda made similar mistakes, but at least they tried to make a collectible hardware device.