|Sonnox’s Oxford Denoiser is a sophisticated plug-in that not only lets
you adjust the noise reduction and threshold (including at multiple
frequencies), but gives you a range of additional parameters to tweak
the processed audio.|
When you hear the term “audio restoration” (ar), it probably brings to mind repairing clicks and pops on recordings transferred from 78RPM vinyl recordings, or maybe forensic applications like you might see on a CSI episode.
But audio restoration software can actually come in handy for more day-to-day applications, such as reducing computer and drive noise that leaks onto your music tracks, cutting out hum, removing clicks and pops, minimizing hiss or artifacts during a song’s ringout, lowering the background noise in film or video soundtracks, or cleaning up dialog for podcasts. Some products even offer the ability to remove fretboard finger squeaks and other extraneous sounds, without affecting the surrounding audio.
AR software comes in many varieties, from dedicated plug-in suites to standalone editors to noise-reduction features built into audio editors. (For instance, Sony’s SoundForge includes a very robust audio restoration toolset.) Audio restoration products range in price from less than $100 to more than $3,000. Generally speaking, the more expensive the product, the more sophisticated the algorithms, and the more precise the control you get. That said, you can find some great options in the $300–$400 range, such as iZotope RX (which also comes in the more-expensive RX Advanced version).
Allergic to Noise One of the most common reasons people turn to audio restoration software is to reduce broadband noise. As the name implies, it’s noise that’s comprised of a range of frequency components, as opposed to, say, 60Hz hum. Examples of broadband noise would be steady room noise picked up low signal-to-noise ratio recording situation, machine noise, and even the murmur of voices in the background.
Now that most people record DAW tracks at 24-bit resolution, the chances of recording a noisy track are significantly diminished. But one area where noise runs rampant is on recordings made by the built-in mics on video cameras. Not only do cameras often record in noisier 16-bit audio, but since the mics are mounted on the camera, they’re rarely positioned close enough to the intended source to record it with a favorable signal-to-noise ratio.
Virtually every audio restoration package offers denoising software to help reduce broadband noise. To get the best results, you are usually asked to select a short section of the audio containing only noise (if you can find one), so that the program can calculate the frequencies and levels of the noise signal. This setting is sometimes called the “noise profile” or “noise print.” The software then processes your audio like a multiband gate, squashing down the frequencies in the noise profile, but leaving the rest alone. Often you can adjust the global threshold of the noise print (or thresholds for individual frequencies within it), giving you even more control.
When reducing broadband noise, you must be careful not to overdo the reduction. The more you dial in—either by increasing the reduction amount or raising the threshold—the more digital artifacts will be introduced into the signal. This can result in a sort of underwater-sounding effect that’s both unnatural and unpleasant. You have to experiment with the settings, and try to find a happy medium that reduces the noise sufficiently, but doesn’t degrade the fidelity too much. The louder the noise is in relation to the source to begin with, the more difficult this task will be.
Steady-state noises, like an air conditioner, for instance, are much easier to reduce than variable noise (such as a crowd murmur). The needs of your project will determine whether you lean toward more intelligibility with a less natural sound, or vice versa.
If you want to apply denoising to a music track, you’ll have to be even more judicious in the amount of reduction that you dial in, because fidelity is even more important. Be sure to solo and compare the track, both processed and unprocessed. Try slowly backing off your reduction amount until you hear the noise returning, then raise it up just enough to make the noise disappear.
|Fig. 1. iZotope RX allows the user to select and edit audio within user-selected time and frequency ranges.|
Noises Here, Noises There Random noises of short duration are potential gremlins in music recordings. Whether they’re finger squeaks, chair squeaks, or car doors closing outside your window, they can ruin an otherwise well-recorded take. Some AR programs, such as iZotope RX and Algorithmix Renovator, allow you to isolate and remove such noises without discernibly disturbing the surrounding audio.
There are a couple of tools for doing this. First, a spectral editor (see Figure 1) lets you select sections of audio in both time and frequency domains, which makes it a lot easier to isolate a stray noise for attenuation or removal.
I’ve used RX’s Spectral Repair module extensively, and I find its ability to get rid of finger squeaks and many other types of stray noises to be downright uncanny. Like the other processes described in this article, use it with care and try to select only the offending audio from the spectral display, or it can create unnatural-sounding results. But when properly applied, it’s very impressive. I’ve also used it to fix a ringout note with a chair squeak in the middle of it, by setting Spectral Repair to replace that squeak with interpolated audio from before and after it. In other words, based on its calculations of the material that came just before and after the squeak, the software generated a replacement that fit in seamlessly.
I’ve covered examples of just some of the uses for audio restoration software. It can be a very handy part of your DSP arsenal, and, once you own it, you’ll be surprised how often you use it. Mike Levine is a regular contributor to Electronic Musician.