Interestingly enough, CoolEdit Pro shows the proper real bit depth of that f...ing mp3 file, which is 16-bit.
How many
bits per sample are available with
lossy coding? Even at (stereo) 320kbps, ~4.3 bits/sample.
Of course, even lossless entropy code would reduce the file size. And, "joint stereo" can be used for stereo.
Using the
"Codec Preview" facility provided by iZotope Ozone, set to
AAC @ 320kbps (I think it uses joint stereo for all bit-rates; no setting is available), the input and output looks in Bitter looks like this:
Where the
source (16-bit music file from a CD) is on the
left, the
encoded/decoded output from iZotope is on the
right. (Both are the left channel, actually, although the codec is being fed both left and right channels.) Indeed, it returns 32-bit float.
Conveniently, iZotope has a "Solo Codec Artifacts" feature. So, comparing the spectrum of the source and the "codec artifacts":
(Left channel only of source and "codec artifacts,"
BLUE=SOURCE,
PURPLE="
CODEC ARTIFACTS.")
Hmm,
hardly "16 bit" performance, is it? (Although
perceptually is another story...)
(N.B. The spectra were averaged over 10 seconds, so this doesn't properly reflect the dynamic allocation of resources (bits.)
By way of comparison, here is the source compared to the
difference between the
source and
truncation to
4 bits:
And compared to the
difference between the
source and
truncation to
6 bits:
BTW, the iZotope AAC codec (at 320kbps) can code a 1kHz sine down to an amplitude slightly
under -160dBFS before "giving up":
(Output after coding/decoding, i.e. NOT the "codec artifacts" output option.)
Here is a multiband compressor that has a
maximum bit-depth setting for its output of
24 bits (varous dither/noise-shaping modes possible, or if turned off then it will simply truncate):
Here is another multiband compressor that has a
16 bits output option:
As such, Bitter is unreliable, because no one knows what processing was used with the recording. I do not care about computed data that carry no useful new information.
Right tool for the right job.
Bitter is a
bit meter. It ONLY reads the activity of the bits that it receives as input; or in other words, it is used to measure bit activity output by a process.
That is why I offered the example of a old EQ plug-in (32-bit float internally) vs. a newer EQ plug-in (64-bit float internally.) Both return 32-bit float to the host software, so
nothing different to note in Bitter; but looking at the
spectral plots, one is a
mess and the other is
clean (given truncation to 32-bit float from 64-bit float.)
IOW, Bitter won't tell you if your signal has been
"screwed up," audibly or otherwise.
I can think of dozens of possible instances where being able to get a quick bit meter reading of what's being output from a process is very useful indeed. Obviously, it is only
one tool in the box.
For example, I have not used foobar in a while, but I imagine that you could use Bitter after resampling to check whether the resampling process used outputs float or fixed-point.
As for your CoolEdit Pro result of "16 bits" for the "Actual Bit Depth," how is it coming up with this figure? Does it mean that the MP3 was decoded to 16-bit fixed point? (I don't have Adobe Audition installed, let alone CoolEdit Pro)
