Steve wrote:
>It is a regular (weekly) occurrence on the forum for users to have
>unrealistic expectations of "Noise Removal". The problem that I am
>suggesting that we address with the name change is not the
>effectiveness of "noise reduction" but the expectation that it can
>"remove" noise - users are not happy when they have spent a lot of
>time to try and achieve the impossible when they are told that at best
>they can only reduce the noise - better imho that they know from the
>outset that they can only expect a "reduction" in noise and not the
>elimination of noise.
 
+1
 
And accordingly +1 to the proposal to rename the effect.
 
Peter.
 
Peter Sampson
Tel: +44 (0)1625 524 780
Mob: +44 (0)7732 278 299
From: Steve the Fiddle <stevethefiddle@gmail.com>
To: Gale Andrews <gale@audacityteam.org>; audacity-quality <audacity-quality@lists.sourceforge.net>
Sent: Sunday, October 21, 2012 2:46 PM
Subject: Re: [Audacity-quality] Noise Reduction improvement

On 20 October 2012 23:15, Gale Andrews <gale@audacityteam.org> wrote:
>
> | From Steve the Fiddle <stevethefiddle@gmail.com>
> | Sat, 20 Oct 2012 14:30:05 +0100
> | Subject: [Audacity-quality] Noise Reduction improvement
>> Attached is a patch that increases the window size in the Noise
>> Removal effect from 2048 to 4096.
>> In my tests this patch typically allows 3 dB (A-weighted) improvement
>> to the noise removal for the same quality of remaining sound and only
>> about 1% increase to the processing time. This is only a very modest
>> improvement, but an improvement nonetheless. If others can confirm my
>> test results then I think we should go for this change.
>>
>> Note for anyone testing: In most cases the optimum settings will be
>> slightly different with the patched version than with the un-patched
>> version.
>
> I only tried it on Windows 7 (x64, 2.3 GHz, dual core, 6 GB RAM)
> using a track with loud white noise mixed in. Just on that (aural)
> evidence, I would say the 4096 window size reduces the noise
> level very slightly without adding more damage.
>
> The optimum setting using 4096 seemed to require a slightly
> lower "Sensitivity" setting than 2048.
>
> However before testing further, my concern would indeed be
> processing time. On a one hour 48000 Hz stereo track (three
> attempts with each window size), I get a processing time of
> about 4 minutes with 4096 compared compared to about
> 2 minutes 30 seconds with 2048. I assume your tests are on Linux,
> but if my tests are representative this would have to be a control
> in the effect I think.

Yes my tests were on Linux (Debian Squeeze).
For 1 hour stereo high level noise (white noise in left track, pink
noise in right track) 32 bit float, 44.1 kHz, I'm getting:
Window size 2048: 4min 50 seconds
Window size 4096: 5 min 24 seconds
This appears to be a "worse case scenario".

For a real-world test - a stereo music recording a little under 5
minutes duration, 32 bit 44.1 kHz:
Window size 2048: 26 seconds
Window size 4096: 27 seconds

It's a shame that the performance is so much worse on Windows - does
anyone have any ideas why this might be?
If the 60% increase in processing time on Windows is unavoidable, then
the modest quality improvement is not imo worth the extra processing
time.


>
>
>> On a related issue, there is a feeling among most of the forum crew
>> that the Noise Removal effect would be better named "Noise Reduction".
>> We regularly receive posts where users clearly have an unrealistic
>> expectation that "Noise Removal" should be able to "remove" noise,
>> whereas we all know that at best it "reduces" noise. It is unfortunate
>> that "Noise Removal" is such a long standing name, but I think that
>> the proposed renaming is sufficiently close to the original as to not
>> cause confusion for existing users, but may provide a more realistic
>> expectation for new users.
>
> FWIW, I'm 0 on this. I can see the point but I think it's probably
> too late to make this change without implying something has
> changed for the worse in the effect
>
> I only have Goldwave and CoolEditPro to hand, but both have
> "Noise Reduction" effects (without controls for Dolby etc.).

I can't think of any other program that makes such a bold and
unrealistic claim as "Noise Removal" (though "denoiser" is perhaps
quite ambitious)
A round-up of some of the names used for post-processing noise reduction tools:

Sound Forge: Noise Reduction
Voxengo: Redunoise
Reaper: Spectral Subtraction
Wavelab: Denoiser
Virtos: DeNoiser
Sonox: De-noiser
Audition: Noise Reduction
ProTools: N/A
Sonar: N/A (noise gate)
WavePad: Noise Reduction
GWC: Denoise
AVS: Noise Reduction.

It is a regular (weekly) occurrence on the forum for users to have
unrealistic expectations of "Noise Removal". The problem that I am
suggesting that we address with the name change is not the
effectiveness of "noise reduction" but the expectation that it can
"remove" noise - users are not happy when they have spent a lot of
time to try and achieve the impossible when they are told that at best
they can only reduce the noise - better imho that they know from the
outset that they can only expect a "reduction" in noise and not the
elimination of noise.

Steve


>
>
>
>
> Gale
>
> ------------------------------------------------------------------------------
> Everyone hates slow websites. So do we.
> Make your web apps faster with AppDynamics
> Download AppDynamics Lite for free today:
> http://p.sf.net/sfu/appdyn_sfd2d_oct
> _______________________________________________
> Audacity-quality mailing list
> Audacity-quality@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/audacity-quality

------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_sfd2d_oct
_______________________________________________
Audacity-quality mailing list
Audacity-quality@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/audacity-quality