Re: [ARSCLIST] High sample rates for noise removal (was 2GB limit for audio file formats)

[Eric Jacobs <ericj@xxxxxxxxxxxxxxxxxxx>]

Goran Finberg:

> So my point is simply that there´s very little content above 20 khz
> coming off mechanical transducers and due to this a greater sampling
> frequency than 44.1 kHz will not add anything useful for the software > to
chew on.

I beg to differ.  Yes, the noise signal may be signficantly rolled off, but
it is still there, and it is still useful to the DSP algorithm.
Sure, DSP algorithms can work without the higher frequencies, but they
can also make use of them if they are present.

There are other side effects to consider as well.  The mechanical
response of a MC transducer is quite different to a MM when it comes
to mechanical imperfections (scratches, cracks, gouges, etc.).  In
the case of the MC, the noise that is produced not only has much
higher harmonics than the MM, but the fundamental frequency of the
noise is higher as well.  This tends to put a bigger gap between the
audio signal and the noise signal, which also makes removal of the
noise easier.

For some real excitement, it would be fun to take the same record
(which should have some significant noise) and ship it to multiple
participants to do their best transfer and restoration, and then
share results.  We can talk theory and physics all day, but in the
end what really matters is what works.  And the only way to compare
theory and practice is to perform experiments.  Any takers?  Anyone
want to moderate such an experiment?


Eric Jacobs
The Audio Archive