[Gwc-general] correlative noise reduction function for mono records?

[David B. <da...@ja...>]

Here's a noise reduction idea I'm toying with, and plan to write some 
sample code in the next week or so.

On a particularly noisy monophonic record, played with a stereo 
cartridge and listened to on headphones, I can hear that a lot of the 
noise is "stereo", i.e.., not exactly the same in both channels.  The 
desired mono signal though, is in phase and roughly equal in amplitude 
between channels.

Seems to be it'd be possible to obtain several dB of "free" noise 
reduction (no aural side effects) from a mono record, by using that 
stereo information to decorrelate information that isn't equally present 
in both channels.   I can think of several algorithmic ways to approach 
this, one of the simplest being to go through the stereo file sample by 
sample and convert it to mono by taking whichever L/R sample has the 
lowest amplitude.   Another option would be to let the correlation 
determine amplitude, e.g., the more the L/R samples differ, the lower 
output amplitude to use.  Methinks a well tuned algorithm could produce 
very nice results.

This would not help scratches or other noise present in phase and 
equally between channels.

My motivation for looking at something like this is because the existing 
noise reduction methods in GWC are effective, but in tough cases can 
color the sound - for instance by smearing transients except with very 
small FFT window sizes (which have their own sonic problems).  Any way 
to leverage information already there to make it possible to use less 
aggressive settings for the more powerful NR algorithms seems worthwhile.

Dave