Rich Peet <> wrote:
> I never post in 8 bit because of the increased noise. The concept
of
> having the idea make sense is still flying by me. When I convert a
16
> bit file to 8 bit the noise doesn't go down anywhere that I see.=20
All
> I hear is a degraded more noisy file. What am I missing?
>
> By the way, you wont likely even be able to tell what I recorded
in
> atrac when converted to 8 bit. Did you find any defect in my
atrac
> re-recorded frog I posted on this thread?
Hi Pet,
I agree, a 8-bit file is not very nice. Converting a 16 bit file to
8 bit will of course not reduce the noise. What I meant is something
different. Imagine you had a 16 bit recording with a high noise
level (either microphone or environmental noise) that is only 40 dB
below the peak level of the sound you are interested in. A 8-bit
file (having a dynamic range of 42dB) would be sufficient to carry
all the relevant information in that recording. I always normalize
the 16 bit file before I convert it into 8 bit. In this way, all the
relevant signals exceeding the noise floor will be preserved. In
case the original dynamic range was larger than 40 dB, the 16-to-8
bit conversion will lead to additional noise. As long as the
foreground sound is loud enough (and without any silent breaks), the
remaining file will be still acceptable. An example for this effect
would be this chorus of European Marsh frogs I recorded several
years ago on Crete (155 KB download):
http://www.avisoft-saslab.com/sounds/frogs.wav
Similar masking effects take place in the ATRAC encoder, except that
the bit-depth is varied continuously over time and frequency,
depending on the current sound. Therefore, silent breaks between
louder sound elements will not suffer because ATRAC increases the
absolute resolution at soft sections (it uses internally a floating-
point format).
>By the way, you wont likely even be able to tell what I recorded in
>atrac when converted to 8 bit.
In extreme situations as in the file posted by Jeremy I was able to
distinguish between DAT and MD based on 8 bit files. The reason is,
that ATRAC may reduce the bit-depth temporarily below 8 bit. The
resulting quantization noise caused by the ATRAC compression is
still higher than the quantization noise of the 8 bit file.
> Did you find any defect in my atrac re-recorded frog I posted on
this thread?
No, I did not find any significant defects. On the spectrogram I can
only see some very subtile low-pass filter effects (above 12...13
kHz). Most of the energy in these calls (the dominant frequency) is
concentrated at about 3.4 kHz. This relatively low frequency is
treated more accurately by ATRAC. This is caused by the implemented
psycho-acoustic model that tries to maintain the quality of lower
pitched sounds, which are more important for your perception.=20
Best regards,
Raimund Specht=20=20=20=20=20
________________________________________________________________________
________________________________________________________________________
|