NoiseRemoval.cpp

[Paul L]

Thank you for catching that, Edgar. I have not built new configurations yet.

I have updated the Wiki page here, explaining the scary new "advanced" controls which are meant to be hidden from the ultimate users. They exist mainly so alpha testers can figure out the best defaults.

http://wiki.audacityteam.org/wiki/Propo ... ion_effect

[Edgar]

I have updated the Wiki page here
http://wiki.audacityteam.org/wiki/Propo ... ion_effect

I made a couple of trivial edits to the page to make it easier for editors to work on. I am not going to bother to do any spelling/grammar/wording work (I'm the lowliest of the Manual editors doing mainly grunt work) until the page is transferred to the manual - well, unless I see something truly annoying <grin>!

I think that Steve could reasonably be asked to commit the header change(s - I think someone caught one and posted it to –devel).

[Paul L]

I uploaded another patch today! The Noise Reduction feature begs to be combined with my other development, spectral selection. Now when you have a spectral selection box, part 2 of noise reduction applies only to the selected range of frequencies. The box has no effect on step 1 -- statistics are still gathered for all frequencies.

I am trying myself to find a good recipe for improving that tough example of very old phonograph records. I can't say I have. But it prompted me to put this idea into effect.

[Gale Andrews]

But at last today I put up a patch at the developers' board containing many of my ideas for fixes and improvements. No attempt at spectral subtraction.

Now I need to figure out how to make Gale happy with a plan for bug tracking and testing.

I appreciate your work on it. I have not used Audacity Noise Removal since I first found it unusable in 1.2.6 and I have found it unusable ever since (in different ways). By "unusable", I mean I don't want to spend inordinate time tinkering with its settings when I can get a better result in Goldwave with one click.

I re-read the Wiki page and still have my 13 issues listed, corresponding to a distillation of the Wiki content.

Fire away and create your Bugzilla issues. Or if you think you can fix all or most of the issues, track the fixing on the Wiki page and only Bugzilla what you can't fix or don't know how to.



Gale

[Paul L]

Gale, I don't pretend to have written the world's best noise removal effect. But, I was curious to know how Audacity's worked, and found a lot that was "obviously" wrong, and have remedied it. My hope is just to have a better effect now than before.

[waxcylinder]

... I have not used Audacity Noise Removal since I first found it unusable in 1.2.6 and I have found it unusable ever since (in different ways). By "unusable", I mean I don't want to spend inordinate time tinkering with its settings when I can get a better result in Goldwave with one click.

Well I beg to differ, I use it a lot for the removal of Webcast and FM carrier noise - It didn't work well until I used the softer settings that Steve recommended to me (and which I placed in an advisory note in the Manual). The default settings gave clearly noticeable tinkly artefacts - so for my money the default is wrong, but hey ho.

And I for one really appreciate Paul's efforts on trying to improve this for us all - keep up the good work Paul.

Peter.

[Paul L]

Thanks, waxcylinder.

Artifacts can happen still with extreme settings or other misuse.

What I try to fix includes the major problem that noise behind sounds was often reduced much less than noise in pauses.

Also the algorithms for finding noise thresholds from samples would require either long samples, or a compensating Sensitivity setting that you would have to find by trial and error, and would still treat various frequencies unevenly, overattenuating some to avoid artifacts in others.

A minor problem was a short fade in from zero, which might be noticeable if you apply removal to a selection within a track. And there were other minor problems.

Also the Isolate button is properly implemented now, ignoring attack and release and frequency smoothing and reduction, and just giving you the noise full strength.

[Gale Andrews]

... I have not used Audacity Noise Removal since I first found it unusable in 1.2.6 and I have found it unusable ever since (in different ways). By "unusable", I mean I don't want to spend inordinate time tinkering with its settings when I can get a better result in Goldwave with one click.

Well I beg to differ, I use it a lot for the removal of Webcast and FM carrier noise

None of which is hopefully very loud.

If the noise is a bit louder than that, in my experience you just couldn't remove it from quiet music in Audacity without damaging the audio unacceptably. In Goldwave you can, and the default setting works almost as well for any material as any result you can get by playing with settings. I haven't tried the new Audacity effect yet, so I can't really comment. Choosing the best values in the advanced settings that will be hidden to users will be very important.


Gale

[Gale Andrews]

Gale, I don't pretend to have written the world's best noise removal effect. But, I was curious to know how Audacity's worked, and found a lot that was "obviously" wrong, and have remedied it. My hope is just to have a better effect now than before.

I don't doubt that you have improved it, Paul, even before trying it. I greatly appreciated your analysis of what was obviously wrong.

Can I assume

For certain noise profiles there may be edge artifacts in the last 1024 samples at the end of the processed selection, where noise fails to attenuate or even rises

remains unremedied?


Gale