Hi there,
can someone give me some more information about noise removal with the noise gate than I find on http://wiki.audacityteam.org/index.php?title=Noise_Removal?
In my dissertation, I want to explain how I got rid off the noise in my audio files, but simply saying that I eliminated lower frequency bands with a noise gate is not enough!
How is it done exactly, does Audacity use a Fourier Transform?
Cheers!
“Noise Gate” and “Noise Removal” are two different effects.
Both are described in the page that you posted the link for.
There is also a link on that page to a “more technical explanation of Noise Removal”.
The “Noise Gate” effect does not use FFT. It simply reduces the level of the audio when the input falls below the threshold.
Hm…right. The sentence “Audacity’s Noise Removal effect is in essence a multi-band digital noise gate” confuses me a bit, but I do mean the section which is described there, which is Noise removal and not Noise gate if I’m correct.
So…I have a noise sample which serves as the prototype for the background noise in the wavfiles the noise gate is applied to. The noise floor in each of its frequency bands is calculated and used as a threshold for a bank of noise gates. If that noise gate is now applied to speech, does it filter out all sound below that threshold, even if there’s background noise when someone is talking? Or does it only filter out noise when there are no other signals apart from noise?
Sorry, I don’t really get the explanation in the wiki!
Yes, but we use the term “Noise profile” rather than “prototype”.
Yes, that’s about the size of it.
This is where it all gets a bit complicated. In earlier versions of Audacity, the gate would only close when the sound level fell below the threshold level(s). This took the assumption that when there is a significant signal level, any low level noise will be masked by the signal. The advantage of this approach is that it causes minimal damage to the audio that you want to keep. The disadvantage is that if the noise level is relatively high then the noise will be removed between words (when applied to speech) but will be apparent when there is talking.
In recent versions of Audacity, a couple of extra sliders have been added. In particular there is now a “Sensitivity” slider. What this does is to adjust the overall threshold level for the gates to deliberately create mis-tracking between “silence” and “signal”. If there is a quiet (noise only) section, the gates will be “shut” and the level of audio passing through the effect will be substantially reduced (how much it is reduced depends on the “Noise Reduction (dB)” setting). With the “Sensitivity” slider at 0 (default) the gate will begin to open as the signal rises above the threshold. However, if the Sensitivity slider is set greater than zero, the gate will remain shut until the signal level is substantially greater than the threshold, so this will remove noise (and some signal) while words are being spoken. In absolute terms, this causes more damage to the signal that you want to keep, but in cases of high noise level it will sometimes produce subjectively better noise removal.
I’ve updated one of the links on that first page that you were looking at, so it now also links to this page: http://wiki.audacityteam.org/wiki/How_Noise_Removal_Works
As a matter of academic interest, the quality of Noise Removal can be improved by increasing the FFT size, but this is at the expense of slower processing. To change the FFT size requires modifying the code so it is not a user setting. An FFT size of 2048 was selected as a balance between reasonable quality and reasonable processing time.
Thanks a lot for the great explanation!