For the rest of the time when we’re not trying to introduce some form of hiss or distortion to make our tracks sound like they were recorded to tape sometime in the 1960s, it seems like we’re trying to get rid of it. In particular, location and field recordings are often particularly problematic: as soon as you start recording outside of your nicely treated acoustic environment, acceptably quiet recordings become harder and harder to achieve.
While this may not be a problem for a lot of music production where, either the noise will be masked by other sounds, or the noise may be an integral part of the production and the reason you set foot outside of the studio in the first place. Where background noise, hum and hiss can really cause problems though, is in location recording of interviews and dialogue for broadcast, podcasts etc.
This post will cover some of the ways you can use the tools of the modern audio editor to try and reduce the impact of noise on your recordings. I currently use Steinberg’s Wavelab 7 as my audio editor, but you should be able to work in any program that can host VST plugins. Before you start solving problems though, you need to take the time to listen to your recordings and work out what the problems actually are…
Reducing Hum
Ground loops and other issues can often lead to the characteristic 50 or 60 Hz mains hum (depending on where you are) being present in your recordings. The advantage of hum problems is that you generally know what the frequency is; it can sometimes drift around by a few Hz, but it’s quite straightforward to locate and quite narrow band. Buzz is related to hum, but, as there are peculiar distortions involved, can be harder to pin down as some of the frequency content can be enharmonic.
The easiest way to start to deal with hum is by using a parametric eq. For 50 Hz hum, add the first band at 50 Hz with perhaps 12 dB of peaking cut with a very high Q; if the hum moves about a bit, you may need to reduce the Q. As the hum will also be present at the harmonics, you will also need to cut in a similar way (although with smaller gain reductions) at those harmonics. So, if you are in the UK and have 50 Hz hum, cut at 50, 100, 150 & 200 Hz etc. One side effect of using very high Q values is potential phasing problems around the corners, so you will need to listen for artefacts in your signal, to make sure that your not causing more problems than you solve.
Reducing Hiss
Hiss is perhaps the most common of problems, and more difficult to solve than hum, as it can often be very broad-banded, rather than the specific frequencies seen with hum. As such, if you try to solve these issues with eq cuts, you may find you are taking away much more of the signal than you would like. A better solution is to use bespoke broadband noise reduction algorithms. My personal choice is the Sonnox Denoiser (part of their Restore bundle, although the Wavelab versions are cut down from the full plugins) that comes with Wavelab 7, but the freeware Audacity also includes a noise reduction process, which requires the sampling of a noise ‘fingerprint’, which is then applied to the file. If the audio with which you are working has already been edited, you may be able to work with the gaps between words, although noiseprints of >1s often yield the best results.
For most of these processes, their use is always a balance of noise reduction against damaging of the wanted audio: dulling of the high frequencies is a common problem associated with noise reduction processes and, at some point, you just have to make a judgment call as to what degree of noise or quality reduction you are prepared to accept. Perhaps the easiest thing to suggest, at least where the signal to noise ratio is good, is to use an expander with a medium to long attack time to push down the level between words. I prefer to use an expander over a noise gate, as I find gating leaves the results sounding too unnatural. Adding some wide eq cuts at the higher-end of the spectrum can help, but this again can lead to dulling of the wanted sounds. Also, be wary if you use any compression to even out the vocal levels, as this can bring up the noise level between words.
Suppressing other noises
The noises we have covered so far are, in some ways, easier to deal with as the noise is constant, reducing the requirement for complicated detection algorithms. Dealing with sounds such as doors closing, passing traffic, digital clicks, air conditioners switching on, vinyl crackle etc can be much harder to deal with. Some low-end rumbles can be dealt with a high-pass filter or some low-end eq work, but things may not be this simple…
Automating eq’s and filters in your DAW and dealing with these things manually may well be the best solution if you can’t justify, or can’t afford the expense of specific tools such as CEDAR Audio’s products for Pro Tools, BIAS SoundSoap Pro 2 or the Sonnox tools etc. One more affordable VST Plugin for some of those random noises is Waves’s W43 multiband plugin, an emulation of the Dolby Cat43 processor: an invaluable tool for location dialogue editing, which is, at heart, a multiband expander. At the cheaper end of the spectrum, Magix Audio Cleaning Lab or BIAS’ slimmed-down SoundSoap LE 2 or SoundSoap 2 may suit you, but perhaps the most complete package at a price that can be justified outside of the pro-studios is Izotope’s RX2 package.
Summing up
There are many types of noise that can interfere with your recordings and can cause problems with either reducing them, detecting them in the first place, or both. As with many audio processes, there is always a balance to be struck between a problem-solving edit and negatively impacting the wanted audio. One strategy is to chain multiple processes -perhaps a filter, then a noise removal plugin, then expander- to remove noise, rather than try to do everything in one step.
Noise removal and audio restoration is, unfortunately, like so many things in life, the best tools often cost the most; you will get better results with bespoke tools than from using freeware. If you do spend a lot of time editing audio such as location dialogue/interviews for podcasts etc., I do recommend Steinberg’s Wavelab Elements 7, a cut-down version of the full Wavelab, but which still includes the Sonnox tools for noise removal. I would also recommend reading this outstanding article from Sound on Sound on the various noises that can creep into a recording and how to remove it. I also recommend this article (with audio examples); although it focuses on Pro-Tools, the techniques are broadly transferrable; there is a good post on noise removal using Audacity here.
Perhaps the best solution is to try not to record noise in the first place: check your studio set up for ground loops (unplugging laptops and running from the battery can often make a huge difference); record using a pop shield; if you’re recording interviews, stand to shield the hot side of the mic from traffic noise and wind by standing in the way and try and use an appropriate microphone windshield. This, unfortunately, is often easier said than done, and some cleaning up is required, while there area number of almost miracle-working tools available, use your ears to make sure you don’t throw the baby out with the bathwater.
Fynn's House and Audio Tech Notes 