In an earlier article I shared ways to approach restoring damaged sound effects. That dealt with ideas and perspective.
Today’s post will share more direct tips. I’ll explain practical tricks to help you get better results. I’ll start with general tips that can be applied to any restoration app. Toward the end, I’ll include suggestions specifically tailored to iZotope’s RX software.
- Be Aware of Tricky Sounds
Certain sounds are trickier to repair than others. Part of this has to do with the nature of the sound effect. Some of it has to do with the problem audio itself.
Consistent damaged audio is easier to work with. A steady thickness of noise, or pitch of hum is easier to identify, and to extract. The job becomes increasingly difficult the more this alters.
Some examples include cyclical problem sounds, like the swell and ebb of distant surf. HVAC and fluorescent light buzz seem constant, but actually cycle rapidly. Adjusting a microphone creates image shift, which varies the spread of audio abruptly.
Software may compensate for this with “adaptive” settings. Otherwise, expect a challenge.
What about the sound effect itself? Sibilant, swelling, and breathy sounds are more challenging to denoise. Aliasing may become more apparent. Also, drastic denoising affects these sounds more. Breathy sounds become thinner more quickly. Sibilant ones become especially slushy.
Sound effects with diverse dynamic are hard to denoise. Algorithms struggle to match big spikes of audio. Noise extraction follows a bit too slowly, resulting in a brief breath of noise after steep drops in audio.
Watch for these challenging sounds when you begin your restoration career. Set them aside for later, plan ahead, or tinker with better software.
- Prepare For Restoration Early
I suggested mastering as you record sound effects last year. This is similar.
One of the most difficult aspects of restoration is finding a clean span of problem audio, called a noise profile. The noise profile is a “template” that tells the software what should be removed. The noise profile is a guide that divides the poor audio (the profile) from everything else you want to save.
This span must be separate from the sound effect you want to save. Why? A “clean” noise profile helps you analyze problems more easily. Also, if the sound effect overlaps a noise profile, you may mistakenly remove part of the audio you want to keep.
It’s easy to get caught up capturing sound effects. Often we’ll record rapid-fire takes. That results in scant space between your clips. That means there isn’t much free space to find a decent noise profile afterwards.
It happens when mastering, too. We’ll cut tracks tightly. That can be a problem when (sometimes years later) we try to denoise a sound and can’t find a clean noise profile in the track.
Record noise profiles in advance. You can do this by rolling tone before or after each take. I do this every time my microphone alters position, or when levels change significantly.
Save these profiles. Create presets in your plug-in or app. Save them to disk when mastering. That gives you a snippet you can reference later when mastering similar sounds. Label these clips with location, microphone, etc.:
Noise Profile, Garage at home, D50 at 120 degrees, Distant POV.wav
It also helps to leave a second before or after a sound effect when mastering. These “handles” help grab a good noise profile when repairing in the RX app, later. Then, when your track is repaired, simply chop off those handles.
Record the self noise of your mics in a dead quiet room with different gain settings. Then use this as [a starting point to] denoise.
- Preserve Your Work
Everyone knows it makes sense to backup your work before making big changes. This is important when restoring audio for another reason, too. Why?
Technology evolves. The tools available now didn’t exist five years ago. Denoising algorithms and decrackling tools become more sophisticated over time. Why does this matter?
Well, that rare Himalayan monk chant you’re struggling with today may be tidied up easily with the tools of tomorrow. Yes, audio restoration relies on the skill and ear of the technician. However, technology plays such a huge part of audio restoration that it’s a great idea to backup your damaged sounds.
I copy each file I’m denoising into a mirrored “Dry” or “Raw” folder. I append the file with the issue. For example:
Massive thunder blast_PEAK
Quiet Korean temple evening ambience_HISS
There’s no reason not to restore your audio now, of course. Sound effects are meant to be used, not buried in an archive. However, a backup is helpful to revisit your superstar recordings when new tools emerge.
- Finish Similar Files At Once
I mentioned last week that restoration is done in “passes.” You’ll study small regions of a sound over and over.
As that happens, your ear will begin to “acclimatize” to the work. You’ll zone in on just that clip. That’s natural. You’re listening with incredible scrutiny. Your ear will develop a feel for the particular noise, hum, or the frequencies in the sound. You’ve developed that pass after pass as you adjust and strip away the problem sounds.
I find this kind of focus is incredibly helpful. That’s why it’s best to finish similar files all at once. It’s hard to resume that focus after a break.
- Regain Perspective
While it’s best to finish similar files in batches, it’s a good idea to give your work space before committing it. It’s easy to lose perspective when you’ve been staring into the depths of noise profiles for hours on end. It’s easy to lose track of the scope of the whole file when you’ve been focused for hours yanking tenacious hum from your clips.
Did your restoration work improve the file, bruise it, or suck the life out of it? That’s why it’s best to step back. Ask colleagues for their opinion. I mentioned before that I master in “passes,” often over weeks. It’s important to do this with restoration work, too. Fresh ears spot errors.
Using iZotope RX
Newbies can be overwhelmed with the sheer amount of options in iZotope’s RX software. How do you begin?
The manual is good for facts. That describes the tools. What about context? It’s hard to know how, why, and when you should use the modules. Here are some tips to get you started.
- Denoise: Fetch a Better Profile
Sometimes you find yourself struggling denoising a sound effect for the better part of an hour. You tweak settings, apply, listen, and it’s not quite good enough. You may hear a bit of musical aliasing, or slushiness. You undo your work, adjust, and repeat. It’s never quite right.
The solution? Fetch a better noise profile. In other words, often the sound itself isn’t the problem. Finding an cleaner, fresh stretch of audio from elsewhere in the file helps. That creates a better noise profile. Often the results improve instantly and dramatically.
If denoising isn’t working, start over with a noise profile from another part of the sound.
Don’t give up on noisy files. Find a better noise profile.
- Denoise: Test Fragments
Denoising takes a lot of processing power. It doesn’t happen in realtime. That means restoring entire files at a time adds up. Check your work on a smaller region, instead. Where is the best place to do this?
Test your work on a variety of audio. Choose a region that includes consistent audio, dynamic, and a switch between them. That will test how your noise profile works behind quiet sounds, loud sounds, and areas without sound. Bonus: choosing a small portion also speeds your work.
- Denoise: Split the Difference
When you first begin denoising, you’ll have no reference point. Should you choose conservative settings, or aggressive ones?
I begin denoising aggressively, listen to results, then reduce. So, I’ll start the Denoise Reduction slider at 12 (for example). I use that as a baseline. It may sound fine. However, less reduction is always better. That taxes the sound file less, and retains more of its original character. So, I’ll split the difference. I undo the processing, then try it again at 6. I’ll repeat, reducing or increasing the denoising until I find something suitable (to 9 if I need more denoising, or 3, if I require less).
During this process, I listen for errors being introduced, and gradually diminish the processing until I reach a balance.
You could work in the other direction: start conservatively, and gradually increase reduction. Then, when you start to hear aliasing, back off. That works, too. I personally find “working backwards” spots errors more easily.
Whichever you choose, discover the best settings by splitting the difference.
- Denoise: Spot Gating
One of the sneakiest denoising pitfalls is gating. That’s when bursts or breaths of noise linger after sharp drops in volume. It’s almost as if denoising isn’t fast enough to sculpt tightly to the transients of your sound effect. It’s common in sounds with short, sharp attacks: door slams, gunshots, hits, drops, and so on.
Listen carefully to the audio directly after dynamic sounds. Does noise linger longer than it should? Do you hear a faint “chuff” or “exhale” of noise? That’s gating.
It’s easy to fix. Just adjust the “knee” and “release” when denoising. This can make the algorithm work harder, though, and may introduce musical artifacts. It’s a trade off. Find a balance.
Aside from aliasing, gating is the most common denoising error, but one of the hardest to spot.
- Denoise: Leave Noise In
It’s easy to suck the life out of a sound clip. RX is quite powerful. It can extract every grain of noise from a sample. However, once all the noise is stripped away, the remaining sound effect will sound strange. It may become stark.
Removing all noise isn’t the goal when it eviscerates the authenticity of the file. Nobody wants lifeless sound effects.
Instead, leave some noise behind. I don’t mean an amount that overwhelms the original sample, of course. But given the choice between extracting all noise to create a lifeless file, and leaving a hint behind, do the latter. It avoids harsh processing, and may have the effect of smoothing out a clip.
Noise is one of a field recordist’s greatest enemies. Diminish it, but don’t loose sight of the quality of the sound. A “take no prisoners” approach to removing noise may make your sound effect sound unnatural, empty, and soulless.
- Spectral Repair: Copy and Paste Spectrum
Spectral Repair is known for its Attenuate and Repair functions. There’s another trick I like better: copy and paste.
You’re probably familiar with copying and pasting in conventional editing apps. Copying a region of sound saves it to the clipboard. You can paste that same snippet anywhere else as many times as you like. When you do this, your snippet completely replaces what was there before. If you cut instead of copying, it simply removes an entire span of audio.
Spectral Repair works differently. Imagine a sparrow sings briefly during your room tone recording. Let’s say it’s song is at 500 Hz or so. Sure, that’s ruining your recording, but it’s only part of the spectrum. The other 19,500 or so hertz are perfectly fine, right? Why bother losing five seconds just to chop out a single, pesky chirp? Spectral repair allows you to fix just the bird chirp frequencies, and leave the rest alone.
Using Spectral Repair means you save your timeline. There’s no need to remove those five seconds of birdsong and lose the rest of the room tone, too. It may not matter as much for a room tone, but consider a jet pass. If you have to chop out the center of the pass to remove a problem, the fly by will be shortened, and may sound choppy.
Spectral Repair’s Attenuate and Repair functions can do this. This can also be done by copying a portion of the spectrum from elsewhere in the file, and pasting it over top of the problem sound. Only some frequencies are copied and replaced.
How can you do this?
- Use the Time-Frequency Selection tool (CMD R) to select the frequencies around your bird chirp.
- Drag your selection to a clean location elsewhere in the track.
- Copy that section.
- Drag it back over your bird chirp.
- Paste. The bird chirp has now been replaced with clean audio.
Normally traditional copying and pasting this way would sound terrible. The edits would be obvious. However, since you’re only pasting part of the spectrum, the rest of it tends to trick the ear.
Now, you’ll have to be careful. It doesn’t always work. The best results occur when there’s a consistent background. But when it works, it’s gold.
- Find Errors
RX has a cool way to spot errors. When playing a sound, we’re used to listening to every frequency for a certain duration. RX allows us to hear just certain frequencies. How?
- Choose the Time-Frequency selection tool (CMD R).
- Choose “Enable Looping” (CMD L).
- Choose “Play frequency selection.” You’ll hear only those frequencies you’ve chosen. So, you’ll only hear from 50 to 180 Hz, for example.
- Drag the selection.
By dragging the selection you’ll be able to hear audio just in the chosen spectrum. It’s like scrubbing, but only certain frequencies. This helps pinpoint the location of elusive bird chirps, tones, or other problems. Most of the time you can spot this in the spectrogram. This is a good trick when the spectrogram is too muddy to spot errors visually.
- Delete Your Work
And a final tip: if processing diminishes the authenticity of a file, even in the slightest, delete the entire track.
It can be hard to tell. Perhaps a sound is partially repaired, the most you can manage. Maybe it’s fully repaired, but feels lifeless. Neither of those produces exciting sound effects.
It’s painful to wrestle with a sound effect, and then abandon your work. It’s important to do, though. You won’t edit confidently with a weak library.
Polishing Your Sound Library
Audio restoration is complicated, nuanced work. It’s a process of trial and error. It requires firm objectivity, and patience.
For all the effort and time it takes, it really is worth it. Audio restoration will rescue your precious field recordings. It will add that little bit of extra polish to make your sound library shine.
Tweet Follow @paulvirostek
To stay in touch, receive free updates by email newsletter or RSS feed. | Follow on SoundCloud