Noise Reduction

Gale 25Jan14: All the strikeouts must be removed from publicly visible content.

• Peter 26Jan15: Done - they remain for now but commented out in ednotes - they can probably be deleted. Removed the P1.

ToDo-1 Bill 13Dec2014: Just starting on this. I know the dialog image is wrong (we don't have a final one for Noise Reduction yet). Some of the control descriptions are little more than place-holders.

• Peter 13Dec14: Thanks for starting work on this Bill. I need to actually rename "Noise Removal" to "Noise Reduction" with a "Move" in order to leave a redirect behind for any external (normally Forum links). I will need to move this page to a temporary location, do the rename and then copy the contents of this page back there.
• Peter 13Dec14 a little later: Actually on rethinking this (and after discussing it with the very sensible Mrs.S) I've decided it will be better to leave the old "Noise Removal " page there with a note saying it has been superseded and with a link to this page. That way it will be perfectly clear, to the occasional reader that lands on that page directed from an old Forum thread, that Noise Removal has been replaced - it's worth the extra click that the occasional reader will encounter. Now I can go to bed and sleep well not worrying about this :-))
• Peter 14Dec14: I think I've fixed all structural changes in the Manual that relate to this change. Wiki still to be looked at, care needed there as the Wiki relates always to the current released version. I will try to find some time tomorrow to work on that.

Noise Reduction can reduce constant background sounds such as hum, whistle, whine, buzz, and "hiss". It is not suitable for individual clicks and pops, or irregular background noise such as from traffic or an audience.

To use Noise Reduction, you need a region in the waveform that contains only the noise you want to remove.

Be aware that it may be impossible to get a satisfactory removal when the noise is very loud, when the noise is variable, when the music or speech is not much louder than the noise or when the noise frequencies are very similar to those of the music or speech.

Accessed by: Effect > Noise Reduction...

Step 1 - Get Noise Profile

This first step teaches Audacity about the noise you want to remove by identifying the noise floor of the different frequencies comprising the noise.

1. Select a region of the waveform which contains only noise. A minimum of 2048 samples (0.05 sec. at 44100 Hz sample rate) is needed. Longer is better. If there are very different types of noise in different places in the track, they are best dealt with by grabbing the profile for the first type, reducing the noise for it, then grabbing the profile of the next type of noise and reducing that.

Note that duplicating a very short sample of noise in order to get a longer sample for the noise profile will not help. The effect is calculating the statistics of the noise. Examining repeated samples of the same noise will not change the statistics.

1. Click Effect > Noise Reduction
2. Click Get Noise Profile

Step 2 - Reduce the Noise

1. Select the region of waveform which you want to reduce the noise in, then set the Noise Reduction parameters. This is often best done by trial and error, adjusting the sliders and using the Preview button to listen to a few seconds of audio after noise reduction. Listening to the Residue can also be useful in determining how much damage is being done to the desired (non-noise) sound.
2. Click Effect > Noise Reduction:
   • Noise Reduction (dB): Controls the amount of volume reduction to be applied to the noise. Use the lowest value that reduces the noise to an acceptable level. Higher values than necessary may make the noise even quieter, but will result in damage to the audio that remains.
   • Sensitivity: Controls how much of the audio will be considered as noise. Greater sensitivity means that more noise will be removed, possibly at the expense of some damage to the desired signal. Set this to the lowest value that achieves effective noise removal without the introduction of musical noise.
   • Attack time (secs): How quickly noise removal reacts to the onset of sound that is above the noise threshold. Values greater than 0.02 sec. are rarely needed or useful. A setting of 0 often works best.
   • Release time (secs): How quickly noise removal reduces its action after a loud sound ends. A range of 0.1 to 0.2 sec. corresponds to the generally-accepted times for temporal masking. Longer times may work better if your audio has a slow fade-out.
   • Frequency Smoothing (Hz): This control should be left a 0 except in extraordinary circumstances. If you are still hearing musical noise in the noise-reduced audio, and you have set the Sensitivity as high as possible without damaging the desired audio, try setting this control to a value between 25 and 100.
   • Noise:
     • Remove: Select this option to remove the noise from the selection.
     • Isolate: Select this option to keep the noise, making it more prominent by reducing the other sounds in the selection.
       If you hear tinkly artifacts in the non-noise, this may suggest Sensitivity is set too high. Currently, "Isolate" works best with Frequency Smoothing and Attack/decay time set to zero.

     • Residue: Select this option to hear the sound that is removed. This is useful for finding the optimum settings that do not damage the audio. If you can hear recognizable bits of the desired sound in the residue, it is likely that you have set Noise Reduction too high or Sensitivity too low.

ToDo-1 Steve 09Jan15: Assuming that Spectral Selection for Noise Reduction is to be released in Audacity 2.1, I think we need more about it here. It is a major departure from how noise reduction has previously worked in Audacity (it just caught me out because I was unaware of the feature). I think there should probably be an image showing Noise Reduction being used in the Spectrogram or Spectrogram (log f) view. This section should probably go higher up the page so that it is less easy to miss.

The Noise Reduction effect supports Spectral Selection in Step 2. If a spectral selection exists noise will be treated only in the frequency range specified by the spectral selection. For example, if you wanted to leave the frequencies below 150 Hz untreated you could make a spectral selection from 150 Hz to the Nyquist frequency of the track then click Effect > Noise Reduction. When you click OK to reduce the noise the frequencies below 150 Hz will not be treated. Note that during Step 1 (getting the noise profile), any spectral selection is ignored. That is, the noise profile always includes the entire frequency range.

"Musical Noise:" This typically consists of random bursts of very short tones at random frequencies, sometimes called "bird song" or "tinkly-bells". This is because most of the noise is being suppressed most of the time, but because the noise floor has been under-estimated (Sensitivity is set too low, or the noise profile was not representative of the noise throughout the track), random noise bursts remain, but now localized both in time and in frequency. This can be even more objectionable that the original noise.

Peter 26Jan15: ToDo-2 AS the original author of this section I no longer believe that we need this. It related to the default settings in the Noise Removal that we had in 2.0.6 and earlier which were a little too aggressive for light noise. In testing Noise Reduction for 2.1.0 I have been working with default settings and these seem fine on removing FM and webcast carrier noise - I get no "tinkly-bells". Accordingly I have commented out this section in this ednote. If others agree then I can just remove this section from the page.

Alternative settings

For heavier noise, the default settings above are recommended, increasing the "Sensitivity" slider if necessary.

For light noise (for example FM hiss, vinyl surface noise, cassette hiss or web-stream carrier noise) the default settings in Noise Removal may be a little too aggressive, possibly introducing musical noise.

• Noise Reduction (dB): Try setting this to about 12 dB. It will still have an appreciable effect on low level noise, but will reduce the risk of musical noise.

*Frequency Smoothing (Hz): Try increasing this a little, typically to around 500 Hz.

• Sensitivity control: Reducing this can further help to make the effect less aggressive, but may require a lot of experimentation for only slightly better results. For a quick result, this slider is usually best left at the default 6.00.

For critical work it is best to experiment for optimum settings; see this page in the Wiki for further advice on using the settings for this effect.

Tips

After creating a noise profile, CTRL + R or Effect > Repeat Noise Reduction will apply Noise Reduction.

Removing noise usually results in some distortion. This is normal and there's nothing you can do about it. When there's only a little bit of noise, and the signal (that is, the voice or the music or whatever) is much louder than the noise, this effect works well and there's very little audible distortion. Unfortunately, when the noise is very loud, variable, or when the signal is not much louder than the noise the result may be too distorted.

If you are still having problems the following tips may help:

• Select  Residue from the Noise: options, click Preview, listening for traces of the desired signal. You can usually reduce the amount of desired sound in the residue by decreasing Sensitivity or decreasing Noise Reduction. Decreasing Sensitivity increases the chance of their being musical noise in the result. Decreasing Noise Reduction will increase the amount of noise in the result. When you have found settings of Sensitivity and Noise Reduction that appear to do the least damage to the desired sound (based on there being little of that sound in the residue), click the  Reduce radio button then click OK to apply the effect.
• Do any click removal, compression or other effects after doing noise reduction, not before. It works best as close to the source of the noise as possible. One exception to this rule is applying the optional Notch Filter... to remove hum or whistle before tackling the hiss.

Bill 13Dec2014: I disagree slightly - click removal should be done first, as the frequency content of the clicks may confuse the noise reduction algorithm. I agree that notching should be done before noise reduction.

Bill 13Dec2014: I'm not sure the above advice is needed now with the improved effect.

• Peter 26Jan15: I'm minded to agree with Bill here - and suggest we remove this particular "tip".

  James wrote by email 8Jan15: "I think the tip in the manual about duplicating the track will also become redundant, as I think the 'noise reduction' parameter should be having that effect, when the effect is 'fixed'."

  • Steve 26Jan15: I agree

*Duplicate your source track before you apply noise reduction, and adjust the relative volume of the two tracks to get the best sound quality.

• If your problem is mains hum or a high-pitched whistle, the use of a Notch Filter... may help. This page on the Audacity wiki outlines the steps. Apply this effect before applying Noise Reduction.

Peter 09Aug14: ToDo James wrote by email yesterday "When the effect is fixed the tip in the manual about using a notch filter (for mains hum or whistle) first won't be needed as the noise reduction will/should behave perfectly when used as a notch filter. It's quite clear from my tests that it not doing so is the bug Paul talks about."

• Bill 13Dec2014: I disagree with James. For mains hum the frequency discrimination of the noise reduction effect is too coarse (about 22 Hz per band). Notch filters work much better on mains hum.

When using Noise Reduction in Chains the profile is captured as below. If a Noise Profile exists, that Noise Profile will be used. It is often best to capture a suitable Noise Profile before running a Chain. If a Noise Profile does not exist: If the Chain is applied to the current project, the current selection is used to create the Noise Profile. Therefore, other effect commands in the Chain will also only apply to that selection. If an export command is added, the entire file will be exported. If the Chain is applied to files, the first file (all of it) is used to create the Noise Profile. It may be useful to prepare a file containing a suitable Noise Profile and name it so that it is alphabetically the first file of those to be run in the Chain.

For advanced tips please see the Noise Reduction page in the Audacity Wiki.

ToDo-1 Bill 13Dec2014: We mustn't forget to update that wiki page!

• Peter 14Dec14: absolutely Bill, it's on my list of things to do. It's mentioned on the 2.1.0 clean-up plan. Care has to be taken as Wiki always relates to current release whereas this development Manual always refers to next release.
  • Peter 15Dec14: I added a new How Noise Reduction Works page and a new Noise Reduction page to the Wiki. The old Noise Removal pages in the Wiki need to stay there until 2.1.0 is released - then after that we can let Mr Deckard deal with them, I've added P1s there so Rick can find them easily ...

    The new Wiki pages require some serious editorial input ...

For a more technical explanation of the algorithm used in Noise Reduction, see How Noise Reduction Works in the Wiki.

Also see Sample workflow for LP digitization which lists some alternative noise reduction software that you can use as part of your audio restoration routine.

Links

|< Index of Effects, Generators and Analyzers

|< Effect Menu