Spectral Subtraction Mode in CNF

[Dan McDonald]

The manual says that the threshold does not operate in spectral subtraction mode of the CNF, but it seems to. If you take a noise sample in SS mode and preview it, the threshold adjusts up and down and there is a noticeable difference. Does that not have an effect on the result itself?

If it is working, how is it different from the attenuation function? I seem to get slightly different results where the attenuation slider is producing artifacts more quickly than adjusting the threshold in spectral subtraction mode, but that could be my imagination.


Dan

[Craig Maier]

Hi Dan,

I am not sure - - - been too long since we created that feature. I suppose that we could have set it up so that the user can "weigh" different frequency bins differently in terms of the degree of spectral subtraction applied. Or, maybe we changed it at some point during the product genesis.

I will try to find out the answer.

Craig

[Craig Maier]

Hi Dan,

It looks like the threshold controls the overall level of the signal which is being subtracted - - - thus the threshold control will have an effect.

[Dan McDonald]

OK - is that different from attenuation, or are they the same thing?

[Craig Maier]

Dan,

I think that they are the same thing. In other words, the two controls are essentially redundant. You can use either the attenuator or the threshold and accomplish the same end result. Raising the attenuator or the threshold will produce the same levels of subtraction. Similarly, lowering either control will reduce the degree of spectral subtraction.

I hope that helps,

Craig

[Dan McDonald]

OK - thanks Craig.

Dan

[mikeebb]

Just a note - bringing up an old thread, I know: I've been playing with spectral subtraction a bit lately and for Really Continuous noise (like tape or FM radio hiss, and sometimes background noise on an LP remaining after a couple rounds of impulse filters) it seems to do a better job for me than playing with the attack/release/etc parameters of the normal CNF. At least, it seems to have less of an effect on the real signal in terms of birdies and similar artifacts. For a well-recorded but slightly hissy cassette it can render the background nearly silent without messing noticeably with the music, based on taking a noise sample from the lead-in then subtracting it from the whole tape.

There does seem to be a difference between the attenuation and threshold adjustments, especially when artifact suppression is on - attentuation has to be (can be) pushed much harder with suppression on but the threshold seems not to be affected as much. With a couple of passes (using Keep Residue during Preview to adjust things to where I was just below hearing any signal in the noise) of higher then low attentuation I was able to make a very hissy cassette of a noisy FM radio broadcast quite listenable -- not completely quiet but the hiss is now barely noticeable. A little EQ didn't hurt, but very little was needed.

-Mike

[rickc]

Mike
That does make sense. The birdies are caused by the quick changes in the filters as they turn on and off as they go over/under the threshold. In spectral subtraction mode the filters are stationary (the spectral subraction part) so there is less birdies.

The attenuation sets the maxumum amount of attenuation as an absolute number (in all modes) and in Spectral subtraction mode the threshold shift does scale the overal gain of the noise sample. It would seem that they would be equavilent but they are not because of the fact that the max attenuation is absolute and not a a maximum delta.