----- Original Message ----- 
From: "Alyssa Ryvers" <[log in to unmask]>
> On 21-Feb-06, at 9:52 PM, Dave Bradley wrote:
> >
> > First off, you can upsample a 16-bit file to 1 24-bit file with 
> > absolutely no increase in noise or relative distortion. If you find 
> > elsewise, then you are either doing it wrong, or have really poor 
> > software.
> Distortion is a change in the sound. You are adding information that is 
> not there, hence, it is a distortion. "Cleaning up" the sound, is also 
> a distortion, of course. Everything adds distortion, starting at the 
> microphone, so life is a question of minimizing the distortions, and 
> being mindful of which and how you are adding...Personally, if I were 
> going to be doing my processing at 24 bit (which I wasn't commenting on 
> in my last post), I would transfer it at 24...I'm not so sure I would 
> be upres-ing to 24 from a 16 bit; I would have to think about it quite 
> a bit further, and would bet it had something to do with circumstances 
> being that I didn't have hold of the original, and there were very 
> special circumstances to warrant the upresing and dithering in addition 
> to the processing.
Now, I'm still not sure I totally understand this, but...

If the sample rate stays the same, I can't see any way that increasing
the bit rate could have any effect at all. If the original waveform
(the signal) is a series of steps 1/44100 second long, expressed as
one of 65,536 values...then wouldn't 24-bit smapling produce the
exact same waveform? Even though there are now going to be 16,777,216
possible values for the steps...their values have already been
established during the 16-bit sampling...and sampling the digital
file at the same sample rate will just mean that the "machinery"
finds one step every time it takes a sample, which will just repeat
the old values. Am I right here?

But (and this is again as I understand it) changing the rate at
which samples are taken (from a digital original, not the analog
ur-original) will mean that the "machinery" sees values of
steps which are 1/44100 second long...whenever it "takes a look"
at the waveform. Obviously (to me) if the new rate is an integer
multiple of the old one, it will take more samples...but of the
same steps (i.e. if you took a sample every 1/88200 of a second,
you would get two samples of each step, both with the same value).
And (again my interpretation) if the new rate wasn't mathematically
related to the old one, it would take different numbers of samples
of the separate steps, more or less depending om "where it landed"
so that the result would be (it seems to me) lengthening the steps
from which it took more samples.

I suspect all of this would be inaudible to listeners, though...

Steven C. Barr