Rinaldo Bernardasci wrote:

Dear Bob,

Are 24 Bit and higher sampling rates really better than 16/44.1?
  
Definitely... at least for origination. Maybe not that audible for destination. What I mean is that the higher the resolution of your source, the more generations of calculations you can go through without adding perceptible distortion or obvious loss of resolution. 

But then you wrote:

If performed under the same conditions, a recording done at 24-Bit and with a higher sample rate does sound better than one done at 16/44.1.

I don't want to get into an argument over that one, because the variables include the D/A converter itself, and the filtering within the converter. In other words, is it the medium or the message? According to recent informal (but blind) experiments, when done under the same conditions, the differences tend to go away. This points out that if our D/A converters are improved, the differences may go away completely. In recent experiments, when we took a 96 K/24 bit source, downsampled it to 44 using the world's best SRC and dithered it to 16 and then UPSAMPLED it back to 9624, again using the world's best SRC-----so the two can be compared at the same sample rate into the same DAC---the differences seem to disappear or become much smaller. So far we've performed this test blind by myself and by a test group of extremely respected audiophiles and we have not yet been able to prove there is an audible difference! If anyone else would like to perform this same test and comes up with positive results (audible differences) blind, please contact me.

But how does the 24/* sound when downsampled and requantized to 16/44.1? Does it really sound better than recording done at 16/44.1?
  
It can. Since you have choice over the dither. And if you have to perform any calculations (e.g. EQ or other processing), then you must dither at the end, and you should work at the longer wordlength to reduce cumulative degradation. It's one thing to say that you cannot hear a loss in one generation, but what about cumulative losses....

Hmm...Well, I was a bit shocked recently to find that the converted recording doesn't really sound that better - it might not even sound better (!), making it quite difficult to justify the extra disk space, dsp, etc.

Well certainly, downsampling and requantization are potentially debilitating dangerous processes, but I perform them all the time in my work. They can be, or they can be your friends, if done properly and conservatively.

Some plug-ins even perform up/downsampling and requantization on their own (depending on the system used this might be beneficial). Are these instruments already compromising the signal because of the up/downsampling and requantization?

Sometimes, usually not. The advantages of upsampling at the beginning of the chain and then downsampling at the end represent a series of tradeoffs. Upsampling prior to compressing and/or other processing with 2 or 3 such processes in a row----is better. But if you upsample and downsample too many times in a row it takes its toll and the losses are more of a compromise than the processing.

Is there a way to keep more resolution during these steps? Maybe using a special tool? (like Saracon?)

Saracon (a product by Weiss) I've found is the best or one of the best, most transparent for this purpose. I've also used Voxengo R8Brain Pro and the SRCs in the higher end workstations, and sometimes it's hard to hear a degradation, sometimes it's obvious. I tend to stick with Saracon to be safe. It measures textbook perfect. You can compare measured results of SRCs on this website.

Being very discerning with plug-ins?> Or consider doing everything at 16/44.1 because the degradation incurred is unavoidable?>   

Not necessarily so. Digital compression adds aliasing products if not performed at the higher rate. So upsampling/downsampling and putting processing IN BETWEEN is less of an audible compromise than introducing the distortion. Furthermore, to my ears, Saracon upsampling is probably completely transparent, I have no complaints about it. The biggest loss is in the downsampling, and if performed only once at the end of the chain, I feel it is the best solution. There is no such thing as a free lunch, you have to pay for any decision you make.

Best wishes,

Bob