How to remove "wind" sound ?

2005-12-25 18:05:54

Hello.

I have a video (encoded from VHS) with a very annoyng WIND background sound in all the external takes.

Please give any directions where/what is the best way to remove (or minimize) it ?
( obs: I still use my old Cool Edit Pro 2.1 )

Thank you
Regards
Marcio

How to remove "wind" sound ?

Reply #1 – 2005-12-25 20:53:29

I'm not a pro, but here's my 5 cents worth of advice -

If the wind noise is a mostly constant sound, you should be able to minimize it greatly. The more variable it is, the more difficult it will be to remove.

First thing I would do is capture some noise reduction profiles. Find some segments where the noise is completely isolated for preferably several seconds. Select / highlight the area (of noise only) and go to Effects > Noise Reduction > Capture Noise Reduction Profile. Save several different examples to experiment with.

Then with a whole area selected where you want noise reduction applied, go to Effects  > Noise Reduction > Noise Reduction...
Experiment with the various profiles you saved (select "load from file") and settings. After you apply noise reduction, listen to a good portion of the file and decide if the reduction was enough BUT without destroying any (or too much) of the remaining audio.

I mainly just work with different profiles and only adjust the precentage of the Noise Reduction Level, as I'm not familiar with the other settings. I usually get pretty good results like that. Maybe someone else will have more advanced advice regarding settings adjustment. Or maybe Google would find some advanced tutorials about Cool Edit Noise Reduction...

Good luck!

How to remove "wind" sound ?

Reply #2 – 2005-12-25 21:20:09

Thank you Cosmo. I will go deeper on that.

How to remove "wind" sound ?

Reply #3 – 2005-12-25 23:53:25

If the recording picked up real wind, the sound probably varies constantly so the noise reduction transform is unlikely to be very satisfactory. I will however discuss how to get the best results from NR in case that will be useful.

Using selective filters may be helpful if significant amounts of the noise are restricted to one or several relatively narrow frequency bands.

A third approach is to just filter everything beyond a certain frequency if that can reduces the problem sound(s) and still retain enough of the audio you want. This can involve either high cut or low cut filters, or both.

If the recording is valuable enough, very local efforts, virtually millisecond to millisecond, may be undertaken to attack individual instances of the sounds you don't want, using any, or any combination of, these three methods. Unfortunately, even with intensive efforts, if what you don't want does vary significantly over time, it may not be possible to get results that really please you.

Work in single waveform view for all of these.

NOISE REDUCTION
Noise reduction use starts with an adequate noise sample. This means a selection of audio with the noise you don't want and none of what you do want. Whether or not that exists in your recording is perhaps a matter of good or bad fortune after the fact but is something you can often plan on capturing if you know that recording conditions are less than ideal. Listen to your selection at extra high volume to be sure you hear none of the audio you want to keep.

Sometimes it is possible to get a useful, if less than perfect, sample by filtering out what isn't the noise. For instance, applying a high pass filter to remove valuable audio but keep the higher frequency noise before making the noise profile. This particular process is done by first copying the sample to a new file and manipulating it there rather than applying destructive filters to your recording.

For a good profile you will require several seconds of noise with no proper audio. If only shorter intervals exist in your recording, copy and/or duplicate multiple short sections into a new file from which to make your noise profile. (It can be best to adjust the boundaries of your sample to zero crossings before doing the copy step so there are no clicks between the multiple pastes in the new noise-only file.)

The following values for making the noise profile are to assure high resolution, which means a better chance of effecting the noise but not the audio. The noise selection must be long enough to support the high number of snapshots or you can't make a profile.
Snapshots in profile: (12,000 to) 24,000
FFT: 24000   — you may want to read in the Help about FFT size and potential problems with time smear. I find this generally not a problem but situations may vary. In general, values suggested for parameters in the CE help files are very often smaller than optimum.

Then apply the noise profile to the audio:
Noise Reduction Level: from 25 (or sometimes less) to 100
Reduce by: 40
Precision Factor: 9
Smoothing Amount: 1 (to 2)
Transition Width: -10

Optimum results may require variations in all of these parameters but they are a good place to start. You can read about each parameter through the Noise Reduction Help button. Again note my comment about the Help suggested values often not being optimum.

After getting a good noise profile you need to fine tune the reduction parameters so that you remove as much of the noise as possible without destroying the audio you want to keep.  Even with a good noise profile it is possible to destroy your recording by being too heavy handed when applying NR.

Most often this tuning just means adjusting the ‘Noise Reduction Level' slider but can involve any of the other parms. You can approach the best Noise Reduction Level from either side, but I like to start high, listen to the results, then reduce the Noise Reduction Level slider until the amount of applied reduction is inadequate (then, of course, increase again until you have enough). Another approach with almost identical results is to set the Noise Reduction Level to 100 and vary the ‘Reduce by' value to get the best amount of noise removal.

CoolEdit NR has a preview function but that isn't always adequate. Start with the preview, to get in to the general range, but then go outside the NR dialogue. Make a selection in your recording, adjusted to zero crossings, apply NR to just that selection, then listen from before to after the selection. Use the undo button to remove the NR. If necessary, go back into the dialogue to adjust the Noise Reduction Level until you achieve the best results. It's slower and clumsier than Preview, but often more satisfactory.

SELECTIVE FILTERING
The first step is to identify the sound you want to remove. A Parametric Equalizer is a good tool for this, as well as a tool to filter out the sound once you identify it. The first step is to amplify the problem frequency, so you can identify it. For this you select just one center frequency band and manipulate it until you've zeroed in on your problem.

The CoolEdit ‘Hum and Hiss Removal' preset can serve as a starting point to set up an example. You will see it has a 60 Hz Center Frequency and a 1 Hz Constant Width. Change the ‘Constant Width' selection to ‘Constant Q' which, in this case, will give a value of 60. Change that value to 600 to narrow the focus. In any particular case, a Q of 100, or maybe 1000, may be most useful. You may use a lower Q to get into the general frequency area of your problem, than up the Q to identify it more closly.

The slider at the right hand side of the graph says -8dB. Change that to zero as we are currently interested in only one factor (not the Hiss Removal part of this preset). The Low Shelf Cutoff can be set to far left (minimum) and the High Shelf Cutoff to far right (maximum). In any particular instance these values may be irrelevant but you don't want them interfering with your identification process if your unwanted sound is nearer either frequency extreme.

Only band #1 is selected (via the check mark in the preceding box). It's level slider (upper right hand corner) is at -20dB. Change that to +20dB. Now when the center frequency matches your problem sound it will be evident because it is amplified by 20dB. The actual amount of amplification, and the actual Q, best for the detection process is something you may have to determine in each instance, but these values can be reasonable starting points.

To use this setup you select a few seconds of your file that contains the unwanted sound(s). Start with the center frequency as close as your best guess to the frequency of the sound you don't want. Sometimes you can get an idea of the target frequency by using the Frequency Analysis Graph, either actively or passively. Now vary the center frequency in small increments as you preview the selection, using either the slider or the entry box, listening for when the unwanted sound gets greatly louder.

Once you have identified the problem frequency you reverse the level slider to reduce it without effecting other frequencies to too great an extent. You will often need to use a lower Q for correction than for detection, maybe much lower.

This process may need to be repeated several times if there happen to be a number of different problem frequency bands in your recording.

How to remove "wind" sound ?

Reply #4 – 2005-12-31 13:43:39

Thank you very much AndyH-ha. Amazing good information. I need redo my "work" based on that. Once again thank you. HAPPY NEW YEAR !!!