> I'm digitally sampling and cleaning up some old vinyl albums. The tough nut
> is obviously discriminating pops/clicks due to scratches on the record from
> transients in the music. I'm using a package called Diamond Cut 6, which is
> powerful and flexible in terms of the number and type of filters it can
> produce, both fixed and adaptive. It offers both time and frequency domain
> filters, which can adjust their threshold according to some local average of
> high frequency energy, etc. etc. I'm pretty happy with it in general,
> although I'm still finding there's a tradeoff between leaving some clicks in
> the result, or signifiantly changing the spectral characteristics of the
> music.
>
> It strikes me that knowing the period of revolution of the record gives a
> potentially powerful piece of information: At 33.33 RPM, any impulse that
> repeats every 1.8 seconds would be highly likely to be due to a defect in
> the vynyl surface, and not part of the music. My question to the algorithm
> gurus out there is, how would you take advantage of this information to do a
> better job of knowing when to filter aggressivly and when not?
>
> Any thoughts?
>
> J
Reply by JHD●January 31, 20062006-01-31
I'm digitally sampling and cleaning up some old vinyl albums. The tough nut
is obviously discriminating pops/clicks due to scratches on the record from
transients in the music. I'm using a package called Diamond Cut 6, which is
powerful and flexible in terms of the number and type of filters it can
produce, both fixed and adaptive. It offers both time and frequency domain
filters, which can adjust their threshold according to some local average of
high frequency energy, etc. etc. I'm pretty happy with it in general,
although I'm still finding there's a tradeoff between leaving some clicks in
the result, or signifiantly changing the spectral characteristics of the
music.
It strikes me that knowing the period of revolution of the record gives a
potentially powerful piece of information: At 33.33 RPM, any impulse that
repeats every 1.8 seconds would be highly likely to be due to a defect in
the vynyl surface, and not part of the music. My question to the algorithm
gurus out there is, how would you take advantage of this information to do a
better job of knowing when to filter aggressivly and when not?
Any thoughts?
J