lol yes sounds like what I need - dumb, brute & simple
Sounds like a new feature for the PAF
In reality, Noise Removal is a trade-off of several factors. For example:
How much noise to remove?
How much frequency damage?
How much impulse response damage?
Do you just want to remove noise from the quiet parts and allow louder audio to “mask” the noise?
How much “bubbly” artefacts?
How much “swooshing”?
The “best” trade-off depends on the type of music, the type of noise and on personal taste/preference. Unfortunately there is no magic formula. There are however parameters that will usually be in the right ball-park for “average” music and “average” noise, which of course should be the default settings.
The main reason that the old 1.2.6 Noise Removal effect worked better on some audio samples than the 1.3.6 version was that the effect attacked noise within the music far more aggressively. The advantage of the 1.3.6 effect is that it caused far less damage to the louder (above the noise threshold) audio. The point of a “sensitivity” control is that it allows the user to decide where they make this particular trade-off. In most cases I think this should be to aim for least damage (as per the 1.3.6 version), but in some less common cases (the examples where the 1.2.6 effect produced a subjectively better result), the sensitivity threshold can be pushed up.
I find the Audacity 1.3.12ß Noise Removal works MUCH BETTER than before …
in fact very well indeed …
I still think/submit that Audacity could/should fully develop VST capabilities.
Wavosaur has no competition (among freeware anyway) in VST capability.
I have found ways to use REAfir VST noise removal in Audacity but not “fully/properly” -
like REAfir VST works with Wavosaur. I find the new Audacity NR and the REAfir VST NR
about equal - except I like the REAfir options better.
REAfir VST NR (in Wavosaur) takes about the same amount of time to do a movie soundtrack
as the new Audacity NR does; however, REAfir VST NR (in Audacity) takes longer
(on my XP laptops) than the actual length of the soundtrack.
The REAfir VST NR (in Aud & Wavo) does seem to add a fraction of a second of silence
to the beginning of whatever sound sample it processes, however …,
which requires an A/V sync adjustment …
Then you have a treat in store for you.
The Noise Removal effect has just been updated in the source code (I’m not sure if the updated code is available yet in the 1.3.13 alpha version, but if not it is likely to be available very soon. The main changes are:
The “Attack/Decay” slider has been fixed (In Audacity 1.3.12 it was stuck at zero regardless of any setting the user made) allowing any “wooshing” to be reduced.
An additional “Sensitivity” slider has been added - setting this at 0 will produce exactly the same effect as previous versions that do not have this control, but increasing this setting can help remove “stubborn” noise that is audible within the music.
A check box has been added to allow “Noise isolation”.
The overall effect is that it is possible to achieve much better noise removal with less unpleasant side effects.
Full VST capabilities will not be possible without real-time processing ability (and MIDI support), so it’s not likely to happen any time soon. However you have probably noticed that VST support is considerably better than it used to be.
For whatever it is worth I mention that I have only ever used the Noise reduction (dB) slider.
Frequency smoothing was set at 150 and Attack/decay was set at .15 as defaults.
The Attack/delay was really at 0?
Yes, stuck at 0 since version 1.3.8.
It’s amazing that no-one noticed
Now you have the option to “choose” to set it to 0.
The changes bringing in the new sensitivity slider, Remove/Isolate buttons and fixing Attack/Decay were committed on 4th July so yes they are in all 1.3.13 alpha Nightly Builds.
I think only a very small proportion of Noise Removal users ever deliberately adjust attack/decay because it conveys nothing to most of them. Short of calling it “Response Speed” (which then wouldn’t be clear to advanced users), RTFM.
Further to the REAfir/AudNR discussion …
(before release of the new new AudNR with the sensitivity slider …)
and noise reduction in general …
Both create the noise sample from playback of the isolated noise.
REAfir also allows drawing a free hand or “points” noise waveform.
REAfir displays the noise waveform graphically.
Users can move the REAfir waveform up or down
which I guess is similar to the AudNR Noise reduction (dB) slider?
REAfir has an FFT level option. It seems that the higher the FFT level -
the more precise the noise waveform. I suppose the AudNR attack/delay slider
functions like the REAfir FFT level?
I’m not sure what the REAfir equivalent of the AudNR Frequency smoothing is …
Any helpful replies/insights appreciated.
AEN007, If you want support with REAfir, you must ask Cockos, but you may find this page useful.
The fundamental reason REAfir doesn’t work properly with Audacity (as far as I can see) seems to be that Audacity doesn’t support real time effects.
Why not try the Nightly Build and see if grabbing the Signal Profile as well as the Noise Profile improves Audacity noise reduction? If it doesn’t, upload an example WAV up to 10 seconds long comprising noise then signal-with-noise, so that Marco can work on improving the new algorithm.
I have no problems with using REAfir features. I was/am not asking for REAfir support.
There is not one question in my previous post about REAfir.
My previous post was in response to the above post/quote.
The questions in my previous post are about how to better understand the AudNR features perhaps by comparing/contrasting REAfir with AudNR …
I thought that would be easy for an advanced user to do.
Any helpful replies/insights appreciated.
Comparisons between the Audacity Noise Reduction effect and ReaFir are, I think, only relevant to a point. While they apply similar technologies, the design goals are somewhat different. ReaFir aims to be a versatile FFT/dynamics processing tool, whereas Audacity Noise Reduction aims to be a simple to use Noise Reduction tool.
I think that is more like the new “Sensitivity” slider control.
If you mean the FFT “size” parameter in ReaFir, then that does create a similar effect to the Attack/Decay control, but only as a side effect of the FFT size rather than being an independent control. I’m not totally sure here (you would need to check the Audacity code for a definitive answer), but I think that the Audacity Noise Reduction uses a fixed FFT size and the Attack/Decay settings operate directly on the gating function.
I don’t think it has one.
+1. However ReaFir seems to use spectral subtraction and I know Marco has tried that out with some success. I’ve got a bit lost in the detail of whether the latest patch with capture of Signal Profile (i.e. in the Windows Unicode Nightly Build) includes that or not. Perhaps he can clarify if no-one else knows.
I’m afraid “I’m not sure what the REAfir equivalent of the AudNR Frequency smoothing is” looked like one to me.
It’s highly unlikely ReaFir will work in Audacity in the short or medium term. So your help in providing samples that noise remove much better in ReaFir than in Audacity would be much more productive in improving Audacity’s noise removal.
AFAIK ReaFir is more similar to the current AudNR than spectral subtraction. I’ve never used ReaFir myself but, judging from what I saw (on youtube), it behaves like a multispectral noise gate (just like current AudNR).
Implement Reafir in AudNR is (AFAICS) almost as simple as using an EQ-Curve instead of the noise sample.
I think it can do either, depending on which mode you are using: http://www.cockos.com/wiki/index.php/ReaFIR#ReaFIR_Modes
My posting was/is about understanding what AudNR does -
perhaps by using the visual examples that REAfir can provide. Thanks for the link in any case.
I have (once again) started working on my mp3/CD audio collection instead of my avi/MPEG/video collection. More than once I have experienced that REAfir removed noice from a CD rip better than AudNR - which would leave (or create) squeaky/scratchy “artifacts”. AudNR worked as well (or better) than REAfir on video soundtracks. I don’t know how or why this audio/video difference should exist???
I don’t “see” how that can be.
It seems to me that the AudNR dB slider is exactly like moving the REAfir waveform up or down …
Does anyone know or could anyone find out what FFT level AudNR uses?
I use the 16384 (2nd highest) setting (in REAfir) for best results.
It is still now clear to me exactly what the AudNR options/settings do. I can clearly see/understand what the REAfir settings do.
My procedure for using REAfir in Audacity uses the Audacity PREVIEW option - which I guess is what causes REAfir in Audacity to take at least as long as the audio track to do the noise removal.
Please note that if you’re still trying to remove tape, or any kind of hiss, you can’t really do that. “Hiss” is constructed of most if not all audible frequencies in a random fashion. If you successfully got rid of it, there would be no music left. Every musical note would be represented as some component of the hiss.
So the trick is to try and figure out which parts of the show are valuable and which ones aren’t and straight noise sampling and subtraction is not going to do it. Worse, if you do get a particular technique to work on one show, it may very well not work on a different kind of show. I bet the NR in 1.3 which can have problems with music, works very well with spoken voice. I bet one of the rules is this: If the signal rises above the average of the hiss, it must be valuable and the suppression must be reduced. The rest of the adjustments have to do with how accurately the beginning and ending of a spoken word is selected.
Dolby Noise Reduction has been doing a similar trick in hardware for years. The presence of sound masks noise.
The particular NR software in 1.2 had no, in my estimation, adjustment range between space alien sound and no affect at all. I had to white knuckle the mouse to get any sort of useful work with the adjustment just up from dead zero. I thought it was broken for a long time.
I don’t know how to get around this hiss thing. The software is already doing on-the-fly spectrum analysis and comparing the energy blobs to generate notch filters. 1.3 tries to add level contouring to avoid strange voices and pumping. What’s left?
You run the risk of patching the show in such a way that it sounds OK, but doesn’t sound like the original recording. That’s why I’ve been using Steve’s Noise Gate gently and I may not need to mess with the Noise Reduction.
Of course, if the hiss is loud enough, you have no choice.
Yeah, I’ve already contemplated that noise removal must remove some of the actual audio frequencies as well,
but nonetheless, this summer AudNR removed “noise/hiss” very well from my VHS»mpeg dubs.
I also process many Grateful Dead concerts - some audience some soundboard source.
Either might have originally been captured to analog tape; yet (often) REAfir (and sometimes) AudNR remove the “noise/hiss” without leaving any noticeable side effects.
My routine is more complicated because I like to use NCH WavPad (v3.02) Automatic Gain Control as the final step.
What seems like successful noise removal when NR is done before AGC might not sound as good after AGC.
It seems to me that doing NR before AGC is better than doing NR after AGC.
I have to apply NR to a sample (in Audacity); save; then run the AGC and listen …
I like my files encoded at the maximum possible volume that does not create any distortion/clipping …
I find most audio that I do needs bass boost and AGC and maybe noise removal.
If you AGC first, then the NR has a constantly moving target. Not desirable.
Any CD of Dark Side of the Moon should make a good sample for testing noise removal. I heard a radio interview (decades ago) with Roger Waters (who was trained not to spit on the fan). RW said just before PF was ready to release DSotM, Dolby came on the market; so PF ran the final product through one more time to add Dolby (which RW laughingly acknowledges - is NOT the way to do it.) DSotM was released saying “featuring Dolby NR” - but was not recorded using Dolby NR. The “noise” is impossible to miss - especially after applying automatic gain control. Finding a noise sample is a challenge.
As stated previously in this thread, it is better to do NR before AGC; but in some cases (as with DSotM), the noise (seemingly) has to be removed AFTER AGC … I found that REAfir did better that Audacity NR (as almost always seems to happen with CD audio - as stated previously)
Today I found here»
the Nyquist Broadcast Limiter II (RFT-Limiter-II.ny) -
which seems like AGC for Audacity.
Per the instructions here»
I put the RFT-Limiter-II.ny file in the plugins folder
but Audacity does not seem to recognize the plugin.
Any helpful replies/insights appreciated.