19 Jun
2004
19 Jun
'04
9:46 p.m.
Hi Ron,
Do you suggest that 24bit at 44,100 samples/second is
a better choice than 24bit at 48,000 samples/second?
Nope. Don't want to get into that argument! I merely am suggesting that
anyone who says that 44,100 samples per second is quirky, odd, strange,
or evil probably hasn't thought about any of the advantages of using
that sample rate from a theoretical perspective. Both theoretically
and practically, 44,100 is better for some things and worse for others.
Practical (I'm sure that you already know this, but some may not have
though about it):
An important consideration is the actual clock speed of the sound card
(for example, emu10k1's don't run at 44,100 samples/second) and also
the sample rate of already-purchased libraries. Steve Harris advised
that it is best to record at the anticipated release sample rate,
but this needs to be weighed against where one starts from and what
one has to go through. To truly follow Steve's advice would mean to
obtain a card that actually runs at the anticipated release sample rate
and obtain/generate samples that run only at that rate, also. Some
people think that because they select recording at 44,100 samples/sec,
that this is what the card runs at --- some do, some don't.
Is there any truth to the idea that additional
conversions has the potential to introduce noise
because of math errors or whatever?
Oh yes. I think most people would say that there is more than just
a potential --- you probably mean a potential to actually hear it. Still
yes.
A/D convertors and I imagine they have a big impact.
Make sense?
Not knowing what brands, etc., I'm just guessing that you're hearing a lot
of the difference between 16- and 24-bit. A lot of detail is lost at
16-bit. 24-bit is not just 50% better, it's 256 times better (2^8 more
accurate for each sample).
I haven't tested 96,000. In the phrase
"because I do
quite a bit of processing," what do mean by
processing? Are you referring to DSP processing
(reverb), file format or both? If you don't mind,
perhaps you elaborate on the reasoning for this.
In short, "DSP processing."
I do my own extensive processing of audio files for simulating 3-D
spaces. One stage in preparation for this processing (but not the
processing itself) requires the use of double-precision floating
point arithmetic. So I got into the habit of using as precise
a representation as I can manage, meaning 24/96 or 64/96 due to
inaccuracies I've seen. I use some impulse response functions that
have very sharp spikes. These are best represented at as high a
frequency as one can obtain. Admittedly, though, the benefit of
twice as high a sampling rate pales in comparison to using 24-bit
representations rather than 16-bit ones for most purposes. But
for extensive processing (effects calculations) another factor
of two is a good thing --- probably worthwhile despite the need
for downsampling at the end. This is merely to ensure less of a
chance of producing audible errors with long chains of audio
processing.
--------------------------
I'm sure you've read a lot about all this sample rate and bit-width
business as many of us have. A big problem is that everyone is speaking
from their own perspective and leaving out a lot of assumptions that
may not affect others. User beware, I guess....
If you ask the potential user community about which sample rate to
use for the library, my guess is that you'll hear good arguments for
both 48,000 and 44,100 --- weaker requests for 96,000. Gosh, who
knows which of the 4X's is best? It's not as obvious as some might
claim.