Fairlight Audio Clean-Up
⚠ Decided I have better things to do with my time and picked up Izotope RX Standard when it was about 50% off. Leaving this here for posterity. That said, the EQ stuff in Step 4 I still use.
Getting a decent mic has made a big difference to the sound quality of my local audio recordings1 but I've been remiss in the audio aspect of post-production. In part because I have no idea what I'm doing and was spending more time than I had available faffing around with dials and knobs I didn't fully understand.
But I appear to have developed some electrical or signal interference in my recordings and while Garbage-In-Garbage-Out is absolutely a thing, I needed to figure out how to eliminate or at the very least minimise the annoying sound that to me sounds like I have a giant fan next to me, but apparently only I can hear.2
Don't use wireless headphones for audio work. I also had issues with wired Apple earbuds. I had better luck with in-ear earbuds with the soft silicon tips, but circumaural (encloses ear) headsets were best. Learned this the hard way.
So, audio clean-up crash-course time. Will flesh this out more as I learn. Will use audio n00b terms that are in all likelihood wrong. But if you're using this as a guide, firstly, gods help you, secondly, this has got to be slightly better than the other tutorials I saw which were effectively "Dick around with the dials and levels until it sounds right, et voila."
Step 1: Master Volume
I raised/lowered the individual speech track sliders so that Master sits mostly on the mid-to-lower end of 10-15 (yellow).
Step 2: Noise Gate
Apply a Noise Gate. This defaults Threshold -35dB but may have to drop this some to avoid clipping too much voice. For the issue that I was having, -38dB is what worked. I set Attack to 1ms as that gives me better edges.3 There was also some fiddling with Release but honestly, I can't tell what I'm listening for here. This was set to about 78.
This cleans up where there is no speech, and leaves it sounding a little stilted because of the aforementioned edges where the noise is under the speech.
You can also add Compression here by setting the Threshold to just below the peak. Didn't need or do much because of Step 1. Compression will clean up the extremes.
Step 3: Noise Reduction
Apply the Noise Reduction plugin. I used the same settings as for the Noise Gate, jacked Smoothing up to 2. Wet/Dry (raw vs processed) defaults to 80% but I fiddled between 80-100% until it sounded okay. 80% was about right. I found the Auto speech mode produced more clipping, so I used Manual then hit Learn over a decent section of speech. Don't know if this is what you're meant to do. Will figure it out later.
This cleans up the noise behind the speech and makes it sound less clipped/stilted.
Step 4: EQ (optional)
Activate bands 1 and 6 and set to curves up and curves down respectively. Set 2 and 5 to the round one.
Working from 2-5 (or 5-2) one at a time (not sure if there's a reason you should go one way or the other) max out Gain, playback speech while increasing the Frequency dial until the audio sounds the most whistley or the most annoying. Drag Gain back down to between -2 to -10dB. Stop when it sounds alright. Going too far down makes you sound like you're in a hole.
Repeat for other bands.
For the noise issue, once 1-3 was done, EQ didn't do too much.
Step 5: Save changes as a preset
Follow this guide to save the setting for later use.
Step 6: Properly learn audio editing or something
- Some sections of laughter or exclamation can be quite loud. In Fairlight, select Normalise Audio Levels from the clip context menu. This also allows the volume for these to be more consistently set than from dragging the volume control in Cut or the volume line elsewhere.
Although frankly almost anything is better compared to using my earbuds. ↩
And if I can hear it, good chance that other people younger than me can, too. So you know, probably most of my target audience. ↩
I have no idea if 'clipping' or 'edges' are even words I should be using in the audio context, but if I treat sound post like image processing it makes sense to me. ↩