An article from a company that’s active in VoIP, video conferencing, etc and a recording that was ruined because an ambulance and police car drove by made me think about noise cancellation using two microphones instead of complex machine-learning algorithms. Could it be that, if you use two identical microphones with good off-axis rejection characteristics of which the first is directed towards the noise source and the second towards the signal, you can achieve better noise reduction? It turns out you can, but it takes specific types of microphones and software. The good news is that it doesn’t require expensive equipment.
To see if my idea would amount to something I used two identical Deity S-Mic 2 shotgun microphones. These mics are inexpensive but have a professional finish and specifications that are identical to those of the Sennheiser MKH-416 (which is the industry standard) except for its off-axis rejection at 180 degrees (the Sennheiser is slightly better in that area at rejecting noise). Neither of these microphones by themselves, however, were capable of rejecting the penetrating sound of the sirens of the vehicles passing by.
And then I read an article in a printed magazine on sound technology explaining how you can use microphone arrays to reduce noise to zero that made me wonder if you could achieve better noise cancellation if you were to position your main microphone as you always do – towards the source that the signal is supposed to come from – and as much off-axis as you can away from where the most noise is bound to originate.
From the article, I knew I had to have at least two mics with the same specs, the same behaviour and the same rejection characteristics. The S-Mic 2 fitted the bill.
Using two microphones outdoors – one hanging from a boom pole and the other mounted on a microphone stand – is probably not going to be very efficient when you’re recording a short street interview all by yourself. However, if you were to record an interview with the help of a few others or in an environment where you can ensure the equipment won’t suffer damage from passers-by or worse, then it could very well start paying off to set up a two-mic system.
And the reason is that it would save time in post-production and turn out to reduce noise to lower levels.
What you need to make it work is either a copy of Magix Audio’s SpectraLayers Pro or iZotope’s RX 7 Advanced. With the latest version of SpectraLayers Pro, you can cast or mould a layer with the signal of another layer (layers are not identical to channels or tracks – it’s a purely SpectraLayers thing). Casting the noise layer to the signal layer gave me the best results with a signal that gets cleaned up to about 99% of noise. The second-best solution is to create a noise profile from the noise recording and apply that to the signal recording with RX 7 Advanced, using the Spectral Denoise module.
You will have to start with a recording to two separate files. I used Audio Hijack with two entirely separate recording blocks in one workflow, but you can use Logic Pro X for this task – or any other sound editor or DAW – and export each track to its own file. The reason I used Audio Hijack is that it is the easiest to create a stereo file from the same channel, saving another few minutes of post-production time.
With SpectraLayers Pro, I simply copied the entire noise file and pasted it to a layer of its own in the signal file. With RX 7, I opened the Spectral Denoise module in the noise recording file, pressed the Learn button to create the noise profile, switched to the signal file and rendered. For best results, I ran this process twice. With RX 7, the trick is not to preview first, because that may alter the noise profile.
The results were very clear, especially with randomly appearing noises like sirens because you can also just clean out those using this technique with results that are far better than any other method. There’s a Youtube video that shows you what I did and the resulting sound file which can still be improved upon, but which I finished in under two minutes in SpectraLayers.