Higher sampling rates of 88/96/192 kHz can be an advantage if you put a sound through several editing stages, because it will retain more of the original signal and introduce less aliasing. Because of that it's mostly useful when you record something then put several digital effects on it and then sample it down, most preferably including noise shaping dithering. But dithering has another purpose in that it can make a 16 bit output sound more like a 20 bit output. Like Tarekith already demonstrated it can make you hear low dB sounds that you would not have heard in the 16 bit output.
Software can behave vastly different at higher sampling rates, sometimes better sometimes worse:
- Arturia's Moog sounds ALOT different, much clearer, more into the face, grittier
- all NI Kontakt based instruments (Battery, Acoustic Piano) seem to have a broken Reverb algorithm which makes reverbs sound alot more dry
Coming to the 88 kHz source to a 44 kHz destination question:
In a perfect world we would all like to have as high a sample-rate and bitrate as possible, because that always means to come closer to the original analog continous signal (actually it not continous but sampled at quantum level
). But in a real world we don't necessarily have to record at high sample-rates, it's good enough if software calculated internally at higher sample-rates. That means it upsamples the incoming signal, works it's magic/editing on it and downsamples at the end. Guitar Rig is said to do so when you use "High Resolution" mode. According to the manual the sample-rate is doubled then, but I am not yet sure if that means "double the input rate" or if it means "NOT half the input rate". Also the Creative X-Fi uses oversampling for it's Sample-Rate Conversion (SSRC) algorithm:
As you can see they do seem to use even multiples, but that's only half of the truth. I suspect that only the first upsampling stage (44 kHz -> 88 kHz) is an even multiple, which means that every sample is simply dublicated. That way you don't lose any information by the upsampling process, but also you don't gain anything (it's just two times the same information). The second stage does not just multiply the source signal but use some sophisticated algorithm on it to bring it up to 192 kHz. The last stage samples it down to 48 kHz and very most likely it will NOT just divide the 192 kHz sample down by 4 but also use some other algorithm.
Anyway you see that SSRC involves more than just multiplying and dividing. A sophisticated SSRC algorithm will not just divide an 88 kHz signa by 2 to reach a 44 kHz signal. And because of that having 96 kHz to begin with should be an advantage to having 88 kHz, because there is slightly more information in 96 kHz and more source information means better destination information. But that all depends on the algorithm used to downsample.
A good way to "visualize" what happens is to think of resizing a picture, which is effectively the very same downsampling. Look at this comparison:
Obviously the outcome very much depends on how you downsample. Some of those algorithm result in a sharper picture with more detail, but also with more aliasing, while others results in a smoother picture with less detail but also less aliasing. Oftentimes the best way is not to sample directly from the source to the destination size, but use several stages including sharpen/unsharpen in those in-between steps, which is kind of what the X-Fi does.
So chosing a good downsampling algorithm (preferably including noise shaping dithering for the bit-rate conversion part) can be vital if you use higher sample-rates. Personaly I would prefer if more software/plugins used higher sample-rates internally at only those places where it makes sense so that we can save memory and CPU for other things. Whatever, higher bit-rates seem to make more sense with the use of a good algorithm for mastering that noise shapes artefacts into the inaudible spectrum. But I need to investigate further into this to understand the internals of sample-rates in DAWs.
