This post focuses on the process of removing noise that has made its way into your content using digital processing.
While it’s always best to avoid noise in ther first place by recording in a quiet environment, using well grounded equipment, and using balanced cables, it’s sometimes unavoidable that some noise will have crept its way into your recordings. Also, often you’ll receive tracks from another engineer or an artist that have been recorded using less-than-optimal conditions. To illustrate the techniques used in this post, I’ll be using a free, open source program called Audacity (I’m a big fan of FOSS software, but most of the FOSS audio software has a ways to go before it is as effective and as stable as commercial software. I use Audacity in this example mostly because the techniques are more or less the same regardless of what application you use.)
I’ve used several noise reduction plugins and most of the good ones work much the same way as the free noise reduction plugin included in the most recent beta Audacity (1.3.3). So head over there and download that program if you don’t have a plugin for whatever audio suite you use. I’ve used the Waves Restoration bundle, which I don’t own, and it works even better but operates basically the same way.
Most of these sorts of adaptive noise reduction processes are basically multi-band gates that analyze the frequency thresholds of the noise and apply it to the overall signal. They assume that if the threshold of the frequency is passed, the sound must be signal and it opens up the gate. As a result, heavily applied noise reduction can sound funny, usually a bit like the sound is coming from underwater. This results from the gate’s attack opening and closing as it reacts to the transients. To reduce the obviousness of this effect, either pull back the gain reduction or quicken the attack (or if there’s a smoothness feature, you can adjust that).
Cleaning Up Some Dialog
I received a small collection of audio dialog from the Presidential campaign of former Alaskan Senator Mike Gravel that had a good amount of background noise in the dialog. The dialog was also recorded fairly quietly. Boosting the dialog meant boosting the noise, which was somewhat painfully obvious. I used Audacity to reduce the noise and normalize the audio to an acceptable volume.
I loaded up the clip into Audacity and found a small piece of silence where only the noise is present.
Once that pure-noise selection is found, I head to the Effects menu and select ‘Noise Removal’ (this name kinda stinks… there really isn’t such a thing as noise removal – just noise reduction).
Then, I hit ‘Get Noise Profile’.
basically, I fiddle with the settings until I hear something that sounds the most like the source signal with the least noise. For this bit of audio, I found 22 db of reduction, about 600hz for the smoothing, and 6ms for the attack produced the results I like the most.
The computer does its thing when I hit apply and it sounds like…
You can hear the noise return a bit when his voice comes in. It sounds a bit like a pen scrawling as a result of the NR artifacting. There are some occasions where the Noise Reduction artifacts are actually more annoying than the noise was to begin with. In those cases, I generally just skip NR altogether. If it doesn’t make things better, skip it.
Sometimes when NR makes things worse and the noise is isolated into a particular frequency, a notched EQ filter works better… although that affects the source signal’s fidelity a little more.
In this case, I was pretty happy with the results. Pretty easy!