Raimund Specht wrote:
>A typical
>nature recording should always fit into the 96 dB dynamic range of the
>16 bit format. If the conversion is made correctly, the noise floor of
>the original recording will safely mask the (ugly) quantization noise
>of the 16 bit format.
Actually, it's the ugly noise of -truncation- that he's talking
about. Truncation is simply chopping off the lower bits. That sounds
ugly (and is one reason why people hated some of the earlier digital
recordings).
There was a discussion of dither on the pro audio list, and I asked
about "self-dithering" signals, where it seems like the the noise
floor of the recording ought to make truncation no problem without
dither.
The response, from those who know much more than I, was that no, it
won't work optimally. Correctly forumulated dither doesn't mask
quantization noise, it prevents it. I saw charts showing that it took
a very high level of pink noise to mask the artifacts that a much
lower dither noise would have prevented.
Unfortunately, I wasn't able to get a layman-understandable
explanation out of the real engineers and mathematicians (I just play
one at work). But what I did get was a rule:
Always dither any reduction in bit depth. Better safe than grainy!
-Dan Dugan
|