Crafting CTR - A Look into the Making of a Cyber Talk Radio Episode, Part III
Part III: The Spectral Noise Reduction Process
When we started recording Cyber Talk Radio in our downtown San Antonio studio, it was really nice for the first few weeks. Then the city started tearing up the street right outside of our window. The daily construction noise was really bad for our recordings, to say the least. I had to find some new ways to start eliminating this noise.
After some Google sleuthing, I found that Adobe Audition has a very powerful Spectral Noise Reduction tool. This thing is amazing. If you feed it some audio, it will create a spectrogram of it. A spectrogram is a visual representation of your audio, which is extremely detailed. Some musicians have in fact hidden secret IMAGES into their music! For examples of this, check out Venetian Snares’ “Songs about my Cats” where the musician hid, you guessed it, pictures of his cats into the audio, or most recently, the soundtrack to Doom 2016, where the musician hid several hidden messages in the game’s audio.
Anyways, once the completed episode is loaded into Audition, I listen to the episode a second time in full, this time listening specifically to audio issues that the previous noise reduction techniques left behind. This is usually the construction noise from outside, or things like chairs squeaking during dialogue, or a pen tapping on the table, coffee mugs moving around, etc. The easiest sounds to get rid of, surprisingly, are the loud, consistent ones. For example, a car horn with show up in the spectrogram as a horizontal like, becoming less pronounced the farther off it is. This is easy to identlfy visually, and you can simply use one of Audition’s paint brush tools to paint over that horizontal line to remove it. It is more effective to be subtle with this, as if you take too much away, you lose precious bandwidth from the human voice and the audio quality will suffer for it. One of the most familiar sounds I have to remove is the dreaded scissor lift that seems to have taken up a permanent residence across the street from the office. Like the car horn, this is easy to get rid of, as the alternating 2 tones of the scissor lift will show up in the spectrogram as a repeating stairstep. More organic sounds are significantly more difficult to remove in a subtle way, especially if they overlap dialog and share the same tonal range as the human voice. These include coughs, sneezes, and the like. Multitracking takes a lot of this pain away, but sometimes they are still difficult to mask.
As an audio engineer, each new episode is a challenge within itself, and my process refines each time I have to produce another episode. I am constantly teaching myself new tricks and techniques. I hope this 3-part blog series has been helpful or insightful.