CNF experimentation

[quartet87]

Recently I've been using CNF on 96/24 files with an FFT of 8192 with much better results than lower FFT settings. Last week I decided to push the envelope on a fairly well worn LP cut and try the 16384 FFT. Wow! Listening to the residue there was little audio being suppressed but there was plenty of fuzz, rumble, hiss, and digital artifacts being removed. It was even picking up the light scuff during the quieter passages. The filtered audio was very clean, considering the source condition, with just a few and faint digital artifacts left.

My settings are pretty aggressive too: attack at 30, release at 70, attenuation at 100, and raising the threshold eight or nine clicks from the sampled noise floor. No smoothing either.

Has anyone else tried similiar settings with CNF on high rez files and how did you like your results?

[Dan McDonald]

Hi Danny - I think a lot would depend on what kind of audio you're working with and what noise floor you sampled. If you're attenuating at 100%, you'll get some 'real' sound in the residuals, but whether their loss is noticeable or not really would depend on the source material.

Dan

[cromag]

I will confess that the CNF is still mostly voodoo to me -- the one that I least understand (I think).

I did get extreme with it recently, with settings that were pretty close to yours, Danny. I was cleaning up some LPs that were made in the 70s by a company (Silver Crest) that specialized in on-site recording of school bands, etc. The LPs were almost mint condition -- only played a couple of times -- but the recordings themselves had a lot of unusual noise. The noise was not the usual rumble, hiss or pop -- it was more like an oscillating, grinding noise.

Anyway, I turned the CNF way up. I listened to the "residue" and it was clear I was filtering out some music, but when I played the filtered file it sounded great. The filter took out much more noise than music.

[Dan McDonald]

I've done that with some really wrecked cylinders also. Some that were stored in played so much their grooves were really worn down, then stored in a 'u-store' shed in Florida for a number of years. I felt lucky there was anything left of them, but in those kind of conditions, you can always figure that you whatever you get is worth it, because it's either salvage what you can or don't bother with it. In that case, I always err on 'well, that's what I could get' because I ate to see anyone's musical efforts lost. Well, almost anyone's. There are a few people I would prefer had never recorded!

[quartet87]

Of course all my work is on southern gospel LP's, which can range from the traditional quartet with piano to the instrumentation being a small county-style combo.

The settings in my first post yield excellent results on a good condition LP, with little digital artifacts...only light traces here and there. However a fairly well worn LP is a different story. Playing with the attack and release times seem to be the key, aside from the high FFT. An attack of 120 and release of 240 takes out a lot of audio along with noise, however the output sounds good with nice air and no artifacts heard. The most notable loss of audio is in the bass area but it's not severe...the output still sounds ok...it's just the thought of losing that much bottom end. This is where I trouble deciding which result I want to go with.

Where I take my sample is critical. I take 2 seconds' worth, trying to find an area with the least rumble. There's always some light crackle and CNF will wipe that out with little or none artifacts provided I use 16384 FFT.

With so much power comes so many decisions.

[Craig Maier]

In general, we have found that the settings required for the optimal use of the CNF (Continuous Noise Filter) vary wildly depending on the material being dealt with. That is why the CNF has been provided with a high range of adjustability and a large number of adjustment parameters. That allows it to have the ability to deal with extremely variant situations. It certainly is interesting how on some material an attenuation setting of 30 will produce artifacts yet on other material settings of 100 will not. Of course it also depends on the settings of the attack-release parameters and the fft size, and to a smaller degree, smoothing and even to an even smaller degree - % overlap.