Tips on Cleaning Up (or "Fixing") Bad Recordings

[snskid909]

Alright so I was given a job to help with the recorded sound from a video of a piano concert. The audio is from the built in microphones on the video cameras, and they are quite a mess. I'm am in charge of cleaning up the recording and making it sound as good as possible. This is the problem almost any sound engineer will have to face at one time or another. And here are some tips I've picked up along the way. If anyone else has advise for cleaning up poorly recorded audio, please post.

1. Garbage in Garbage out
Ideally you get a clear recording, but this isn't always the case. Just remember that there's only so much you can do to "fix" a recording. But then again that's what the rest of this article is about.

2. Align the tracks
If you have multiple source of the same thing (ie. the audio from two different cameras) find a spike or clap or similar thing in the waveform to align the tracks. And zoom in to get it as best you can. Sure, you recorded from different parts of the room, but don't let that stop you.

3. Normalize
Levels were bad? The first thing to do is normalize. It will make all the audio the same volume... sorta. But doing it first will save you headaches mix it later.

4. To fix the clipping
When the audio goes above 0db on the recording, your screwed... but there are a few things to try that might save your butt.

a. Click removal, repair, and clip fix
These are the effects that most specifically apply to clipping, Click removal and repair are quite good at recovering your short little scratch or pop, however the the audio where this occurs must be fairly short and be sure to reduce the gain prior to using the effect. Clip fix works by interpolating the audio and can work quite well, but not always. It depends how badly you peak.

b. using the best track.
if you have multiple tracks, and only one peaks, you can copy the audio from the good track and paste it over the bad track. Sure you may have a mono output for 2 seconds, but its better than crazy distortion.

c. EQ it .
Sometimes the clipping occurs only on part of the frequency spectrum. Open up the graphic EQ and bring that sucker down. Highlight the clipped portion and use ANALYSE-> PLOT SPECTRUM to help find what frequencies to cut.

d. Draw it in
As bad as it can be sometimes its possible to rewrite the bad audio. Just zoom in super close and select the pencil tool and rewrite the waveform yourself. Use sparingly, but sometimes it works just enough to smooth over that terrible sound. This is destructive editing, so be careful. If you have multiple tracks you can use the other tracks as guide of how to redraw the waveform.

5.The Compressor
Got a voice you need to hear but its drowned out by another? Compress it! Sometimes you cant pull out the annoying "things" but if you lower the dynamic range you can bring everything else down a bit. There are a lot of knobs on this one, but the most important ones are "THRESHOLD" which determines what point the compression sets in, and "RATIO" how much compression takes place. Generally the Threshold works best between -6 and -24db and the ratio between 2.5:1 and 4:1. This is generally better than just boosting the signal. However realize you will still be boosting the noise as well.

6. Noise Removal
Yes it works. No its not magic. Get a few seconds of just noise for the noise profile, and apply noise removal as needed. Its best to use just the minimal amount needed as it will distort your audio slightly. Try not to use if you need the detail of the recording such as during a musical concert or recording.

7. EQ it
Last but definitely not least is equalization. There are some good articles online you can Google on how to EQ. Its possible to work wonders with an EQ but its also possible to destroy a mix, so tread lightly and always compare the EQed mix with the original to make sure you actually are improving the sound. But basically bring up good frequencies and pull down bad ones. EQing is an art and takes time and practice to get it good and get it right.

Got any other tips to save bad audio, post it below and share.

[steve]

My #1 top tip is to make a backup of the original recording before doing anything else.

6. Noise Removal
Yes it works. No its not magic.

very true.
The main thing to take care about with Noise Removal is not to overdo it.
If there is a lot of noise (and there usually is from camcorder recordings) it will almost certainly not be possible to remove all of the noise without also causing significant damage to the sound that you want to keep. Attempting to remove too much noise will often produce a metallic bubbly effect to the sound. Achieving best results is a balance between achieving acceptable noise reduction while not damaging the remaining audio too much. It is often useful to try out the Noise Removal settings on a few short selections of the audio (Ctrl+Z to undo the effect). When you have found settings that you are happy with, apply the effect to the entire track.

More information about Noise Removal in the manual: http://manual.audacityteam.org/man/Noise_Removal

Noise Removal should be performed before using any compression type effects as these can cause the noise level to fluctuate, making Noise Removal far less effective.


Camcorder recordings often have a lot of low frequency noise (hums and thumps) and no useful audio in the very low frequency range. Applying a low bass-cut filter to the track before any other processing can often make a significant improvement to the sound and help other effects to work better. I generally prefer to use the Equalization effect for this.

[snskid909]

Noise Removal should be performed before using any compression type effects as these can cause the noise level to fluctuate, making Noise Removal far less effective.

That is a very good point to make and something I did not know. But it does bring up another question, is it better to run noise removal before or after equalization or does it not really matter?

Thanks

[waxcylinder]

This tutorial workflow for LP cleanup from the manual gives a good recommended processing order: http://manual.audacityteam.org/man/Samp ... gitization

WC

[PGA]

...Ideally you get a clear recording...

By far the best way to "fix" your recordings! When I first started out in audio visual work in the 1980s, my mentor taught me "You cannot make a good recording out of a bad recording; all you can do is make a different bad recording." Digital technology has not invalidated that advice. The secret still is to get it right at point of original capture.

[snskid909]

Wax thanks for the link it was extremely useful. One other thing that you might also want to note here is that for every additional effect that is used, the original signal quality can take a loss. So it is better to EQ only once instead of reapplying the EQ to get better results. Each additional filter degrades the audio a little more and if you go through 5 EQs it can be quite noticeable. If a filter doesn't sound right the first time, your best bet is to undo it (ctrl z) and play with the buttons a little more.

[waxcylinder]

... for every additional effect that is used, the original signal quality can take a loss. So

A good point and worth reminding folk.

When I do my LP captures the only processing I do is to remove clicks and pops, clean the inter-track gaps with fade-in/out/silence and amplification to -3dB. I prefer minimal processing, maximum original signal.

My signal gets a tiny amount of 50Hz, and its multiples, (frome somewher in the chain TT/Cart/pre-amp/external soundcard - bu I can't tell which) but it requires massive amplification to see it all on a null waveform, so I decided just to leave it there. The TT adds a little LF rumble but that causes no problems with the speakers so I leave that there too.

WC

[steve]

for every additional effect that is used, the original signal quality can take a loss.

A good point and worth reminding folk.

While this is certainly true of analogue processing it, with digital processing the losses caused by repeating effects may be negligible (depending on what you are doing).

A good example of where the loss is negligible is the Amplification effect. As a test, take an audio track (32-bit float format) and make a duplicate of it (Ctrl+D). Then apply -6dB amplification to the duplicate, then +6dB amplification. Repeat this 30 times ( >> -6dB >> +6dB >> -6dB >> +6dB >> ...). Then compare the processed track with the original. Can you hear any difference?