Loading in 24 vs 20 bits

Having increased my RAM, I have loaded some organs in 24 bits . I am under the impression the sound quality is somewhat better ( meaning a bit more clarity and detail ( harmonics) in the sound)...
But I wonder wether this impression is real , knowing that J. Zurek's experiments on his site tend to demonstrate there is no perceptible difference between 20 & 24 bits ( no great difference in aliasing and other parasite harmonics, which are in both cases very much reduced compared to 16 bits).
=> Any opinions on the subject ?

Every ear is individual and hears differently. Choose the highest possible option that suits your main memory.

I am a mathematician who has done a lot of work with digital musical technology during my career. I also maintain very good hearing. I would be surprised if many people could hear the very subtle difference between samples using 20 and 24 bit resolution. I think the greatest differences would be evident if you listen to very quiet ranks - which might require turning up HW's volume quite a bit. These are minor effects that you will likely not notice in typical performance.

In other digital systems, you can begin to hear defects in harmonic content when you play many ranks simultaneously. This is because the overall numbers approach or even exceed the total bit accuracy in the system. I have not noticed this in HW because HW uses MUCH more than 24 bits internally to manager the sum of ranks being played.

The remaining effect that you might notice is also related to the number of bits of resolution of each rank - this is the noise floor. Each rank that is recorded in a live environment has noise - hopefully at a very low level compared to the amplitude of the rank itself (the signal.) Signal-to-noise is a very important metric in audio (and many other) systems. As you add notes and ranks, the noise level adds faster than the perceived signal level, increasing the detectability of the noise to your ear. Noise sources include obvious things like hiss and background noises in the building were the pipes sit. However, they also include internal noise from the recording system itself. For this reason, it is preferable to sample pipes so that their signal level is as high as possible in the recording. A higher bit resolution helps with managing this. However, the number of bits can then be reduced when the rank is introduced to HW.

The bottom line: I typically use 16-bit or 20-bit resolution when loading HW sample sets to save cache memory. I can have more organs ready to load quickly in this way and the aural difference is not something that I can detect in normal playing.

Les

Many thanks to you both for your precious observations !

Our auditory memory is surprisingly short (a couple of seconds). People often report subtle differences in audio quality when using different audio settings, cables, sound cards or amps, or even when listening on different moments of the day. Especially the latter also happens to me These differences tend to disappear in blind tests that are level matched (so no volume differences) and are best explained by psychoacoustics.

Interesting responses, and thanks particularly Les. I load in 24 bit where possible, and the difference for me is it seems to be a bit less noisy, that is the noise floor is lower as Les observes. This makes it seem to have better clarity and detail, I think, because the signal to noise level is better. But I also routinely play sets at 16 or 24 bit, and quite enjoy them (like Groton or Little Waldingfield). I hear the extra noise, but it doesn't bother me after the first few seconds.

I don't really hear a difference between 20 and 24 bit samples, even playing very quiet registrations. But that's with 96 kHz sample rate, highest audio processing and model quality settings, and usually with the sampleset's blower noise and the computer fans supplying a low level noise that might cover up an elevated noise floor. I do hear more noise at 16 bits when playing quietly. I just stick to 24 bits because I have enough memory.

I completely agree with all of your remarks :
1) it is very difficult indeed to get a sure vision of what sounds best ( ...or at least , what you prefer best) best ,especially because of the very short auditory memory . ( same problem when ,for instance , you try to choose "blind- eyed" between 2 or 3 different pairs of stereo loud speakers , even when they are linked to the same amplifier allowing immediate switching , and you are listening to the same piece of music) .;
2) It is logic that the sound is clearer when the noise floor is lower ( the other environment sounds eing also low or null ) + same thing in "real life" with a real pipe organ in a real church !

But, netehertheless, my point was that I was surprised to feel the 24 bit was "clearer" ,when J. Zurek recordings of frquencys in 16, 20 & 24 bits tend to demonstrate there is no significative difference between 20 & 24 bits...

Finaly, on the whole, as you all suggested in a way or another, the question gets down to finding a good balance between available computer memory, sound quality and loading time of the sample set...not forgetting that the two last parameters are highly relative and psychological !.....