quote:

Originally posted by uxud Wow, I was just wondering about this. Does recording at 48khz offer any significant benefit over 44.1khz if you're just exporting to mp3, and how much more CPU power does it use?

I'd say no. If there is any gain in quality, it's VERY minimal. 48 kHz was originally introduced to allow varispeeding 44.1 kHz recordings (varispeed is in fact nothing else than changing frequency). If you did that with a "normal" 44.1 kHz recording, you'd start having problems with aliasing. That's what 48 kHz was for, NOT for the mere gain in quality.

So for mp3 or cd, I wouldn't do the effort of switching to 48 kHz.

Now onto the question of superddman...

When we talk about working with 24 bit files, we mean working with files that have been RECORDED in 24 bit. Of course, for that you must have audio hardware that can handle that resolution, but nowadays those things are getting more common (of course, like Aaronoct said, there's still a difference between consumer gear and pro gear).

One difficult area for dithering has always been those internal engine resolutions of the audio programs. First a simple explanation why they work at such high resolutions internally. I'll explain it with a metaphor, so not to scare of most of you

Let's assume your audio level is represented by a money value of 1.0 dollar. Precision is one after the decimal (.1). Now let's say you want your audio level to be 1/4 of full scale. That would be 0.25 dollar. If you retain that .1 resolution, you'll have to round off. Your actual value would be 0.3 dollar. You see here that you already lost some information. And that only with a simple gain change. So, you can only guess how many info you lose with more difficult processes... It might not sound important to you like that, but don't forget a lot of micro information is described in those bits you lose, like reverb tails, spatial info and such.

Now if you take your 1.0$, and your sequencer says : well I see it as 1.000$. It's still one dollar, but with 0.001 resolution. If you would take the 1/4 level again, the result would be 0.250 And here you see you retain more info.

You'll ask, what's the point if you reduce bitdepth at the end anyway. You'll lose the info no matter what. Well, try it with simple calculations, and see what is the most precise result : losing a bit of information on every calculation, or losing a bit of information just on the very end. Consecutive losses will be worse.

A fairly easy way to prove this, is to use an editor that has a bitscope, like wavelab. Take a 16 bit file for example, and do something as easy as a gain change, and you'll see that the bitmeter suddenly shows you have more bits than the original 16.

The problem lies in the engine. A good professional program will work at a high internal bitdepth, but do nothing else. Some programs have the bad habit to work at a high internal resolution too, but dither down to the working bitdepth after each process again. And like we said, dither should only be done once, at the very end.