Ok, I will attempt to describe what a dithering algorythm does, and why it will sometimes help.
Your bit depth comes with a maximum dynamic range that it can represent. For 16 bit, this is about -96dBfs. Few, if any 16 bit converters actually achieve this dynamic range, but generally get close (maybe -90 to -93dBfs). 24 bit converters inherently can do -144dBfs, but in real life, only offer around -110dBfs (the analog circuit of the converter accounts for this "real world" dynamic range...anyway...).
So, when you record with a 24 bit recorder and save at 24 bit, you can represent about 110dB of dynamic range. Cool.
Now, we know that CD's only do 16 bit (96dB) dynamic range. So, there is a problem here. The 24 bit file must be TRUNCATED down the 16 bit. The act of TRUNCATING just cuts off that extra dynamic range. Simply, it is GONE. The LSB (least significant bit) is still holding a LOT of audio information that the ear is sensitive enough to distinquish. As the LSB toggles On and Off, your ear can HEAR the "digital black" (LSB switched off) and audio (LSB switched on). The LSB being ON is going to be usually 6dB louder than digital black. So, depending upon the musical content, the toggling of the LSB will cause a sort of "distortion" sound. Our ears ARE sensitive enough to pick up on this.
So what to do? We certainly don't want this "toggling" of the LSB to happen, because it tends to sound bad.
What a dithering algorhythm does is add a "shaped noise" (shaped noise is basically noise that is out of the senstive region of our hearing, 2-5KHz) that is "mixed" with the audio. Basically, the act of the LSB staying ON because of the added noise means that audio that is technically lower in volume than the dither noise is recorded on the LSB, thus, giving the illusion of more dynamic range. On a meter, the volume down that low is staying the same, but to our ear, which can hear the music well, but not hear the dithering noise added, we are hearing volume variations!
It is a clever approach to increasing the PERCIEVED dynamic range of 16 bit recordings.
So, just think of dithering as noise that we don't hear so well, mixed with the audio that allows audio that is technically lower in volume than the dithering noise to be heard because the LSB is always ON.
The reason I say that many home recordings, and rock mixes don't really NEED dithering is because these kinds of recording generally already have noise that is louder than dither noise! This could be from noisy preamps, amps that has rushing noise recorded, etc....Also, the genre of music may not be very dynamic, and the production style may be quite dense, thus, not leaving any room for us to really hear a "fade out" , which is when the LSB toggling on/off become appearent.
So, if you already have a fairly noisy recording, or a song that doesn't fade out in any kind of way, you probably won't ever hear a benefit to dithering your audio before truncating it to a lower bit depth.
In some recording, yes, dithering a great benefit! Spacious productions, songs that fade out, etc, really do benefit from dithering being applied before truncating the file.
The real test is to try both. Really, if you can't hear the dither working for you, it is probably best NOT to add MORE NOISE to your recording eh? I mean, dithering is simply adding noise, and if you already have enough noise in your recording to keep your LSB "On", why add more noise?
Too often, dithering comes up as some magical "solution" to "grainy, harsh, stale" digital audio. Far too often, it isn't the fault of the LSB toggling on/off that is causing "grainy, harsh, stale" digital audio, it is a BAD WORD CLOCK, or just plain ol' poor converter designs! Just as often, it is just poor production in the first place!
Truthfully, you have plenty of other things to worry about LONG BEFORE you should worry about whether to dither or not. Really, in 99% of home recording and/or Rock music productions, dithering will make little to no noticeable difference.
Hope this clears it up a bit.
