revisiting automating Noise Reduction

[Edgar]

feature.png (40.15 KiB) Viewed 1698 times

I'm not quite sure how this should work in a project with multiple audio tracks. The way I do it now (very ugly) takes the current selection as the Noise Profile, selects everything in the whole project (ugly but benign) then uses that profile on the Effect's list of tracks.

[kozikowski]

My version of that uses the whole track for the Profile, but selects only for sound that does not change. That's a lot more computationally severe, but probably a lot more accurate.

The Disney Cartoon example is a flip-book-based Plot Spectrum. As you watch it, the spectral outlines bounce, pulse and surge with the show... except for those four peaks which never seem to change. This may be a place where Human Pattern Recognition can see this from across the road, but it may take a machine a bit more work.

Correct me, but isn't that how Cellphones work? NOISY NOISY NOISY noisy n o i_____________________.

I guess it's possible that just averaging out the whole world would work, too. No matter how it's done, it has to be dynamic sampling, full bandwidth for the whole show. That's the only way to get any accuracy at all.

Developed properly, it may be the preferred way to run Noise Removal, not the celebrity you keep locked in the attic.


"Go through the entire show and drag-select the one place that has the most noise with the least amount of presentation or desirable content."
That's how it works now. You're joking, right?

Koz

[kozikowski]

I remember another piece of this. There's a sampling law that states no matter how large the base set, a random sampling of Some Magic Number (like 150 or 200 or so) will get you 98% of the way there and further granularity of sampling doesn't move the number. It's applied to voting and elections, but it's a universal law.

So no, you don't need to surgically sample an entire hour show millisecond by millisecond. Random samples of Whatever That Magic Number Is will do it.

I am, of course, dreaming/hallucinating.

Koz

[kozikowski]

Noise Reduction (not Removal) will have to be set up for each track. That's not to say the User has to do that, but the software will have to do it multiplying the calculation time and making the second pass application phase almost impossible.

Automatically mix and render and apply N/R to the resulting track?

I don't see any graceful way to reduce multiple tracks.

Koz

[Paul L]

Not sure I understand the problem you are solving.

[Edgar]

Not sure I understand the problem you are solving.

A long time ago we had this discussion:
http://forum.audacityteam.org/viewtopic ... se+removal
I contend that for the average user by far the most common workflow is:

• 1) Open the Noise R… Effect dialog
  2) Inspect the values, changing as needed but most likely accepting what was previously used
  3) Get Profile
  4) Select entire track
  5) Apply profile

Currently, this means opening the effect dialog twice with a bit of work in between to change the selection from the tiny bit of noise to the entire track. My proposal is to add a feature to the current dialog which would make this something of a one-click operation:

• 1) Open the Noise R… Effect dialog
  2) Inspect the values, changing as needed but most likely accepting what was previously used
  3) Click a single button which causes the current selection to be used as the Noise Profile, dismisses the dialog, (if absolutely necessary, selects the entire track) and uses that Noise Profile on the whole track (it would be nice if we did not need to change the selection but could still clean the entire track).

I've been doing it this way for at least 3 years and I know it has saved me a lot of time.

[Paul L]

I still don't understand the added convenience here. You can just do ctrl-a, ctrl-r if there is only one track in the project. Or click on the track label at left and then ctrl-r.

[Edgar]

I still don't understand the added convenience here. You can just do ctrl-a, ctrl-r if there is only one track in the project. Or click on the track label at left and then ctrl-r.

<ctrl+r> is not very intuitive in this context as it implies "repeat the last effect on the current selection" and the last effect was to use the current selection to Get Noise Profile. Without your pointing this out I would never have discovered this (even though it is in the Manual)! I suppose even Manual Editors need to RTFM <grin>.

[Gale Andrews]

Aside from the undiscoverability/unexpectedness that CTRL + R after taking the profile applies the effect, is there any reason not to make the Noise R... dialogue modeless?

I would argue that after first capturing a new profile, "Repeat Noise Removal" should say "Apply Noise Removal". The first time I discovered the current CTRL + R behaviour I was trying to change the profile I had captured...


Gale

[Edgar]

is there any reason not to make the Noise R... dialogue modeless?

With the possible exception of the user wanting to apply the cleanup to the entire track from which the noise was selected (my gut feeling is that this is a very high percentage of the cases and why I suggested my solution in the first place) it could be quite problematical but, maybe not, if TrackLists can cross Project boundaries.