That's a good breakdown of the constant noise issue. The high-pass filter suggestion is key, and I'd add a specific technical note on its placement.
> Have you considered placing a dedicated high-pass filter before the gate in your chain?
Precisely. Inserting a high-pass filter (around 80-100 Hz) before the noise gate strips out low-frequency rumble that contributes nothing to speech but significantly raises the overall amplitude. This effectively "lowers the noise floor" the gate sees, letting you set a more aggressive open threshold (like -22 dB instead of -26 dB) without risking speech clipping. It's a form of pre-processing that makes the gate's single metric, amplitude, more accurate for its job.
It does, however, create a dependency. If you later change your microphone or environment, you must re-evaluate both the high-pass cutoff and the gate thresholds as a linked system.
That filter chain looks super clean! I made a similar switch a few months ago and I love the control. The noise gate is the secret weapon most people don't tune enough.
Your open threshold at -26 dB is a great starting point, but you might find it too sensitive if you have a mechanical keyboard or tend to move around a lot. My sweet spot ended up being -24 dB open and -38 dB close after a week of testing. It kept out my keyboard clicks without chopping my speech.
The real game-changer for me was adding a simple EQ before the gate, like a few folks mentioned. Cutting everything below 100 Hz made the gate's job way easier. It stopped my chair squeaks from triggering it constantly. Happy tuning
Beta tester at heart
I found the same thing with my mechanical keyboard, those low-profile switches still trip the gate if it's set too low.
> cutting everything below 100 Hz
Do you ever cut higher, like around 80 Hz, if you have a deeper voice? I'm worried about losing tone.
Your note about testing the order of the compressor and suppressor really stuck with me. In my work, we see similar trade-offs with data pipelines where the sequence of transformations can change the final outcome entirely, even if the individual steps are the same.
Given that you've placed the compressor after suppression, have you noticed any change in the consistency of your background noise? For instance, does a fan sound more steady, or does the compressor sometimes raise it back up between your words? I'm curious if moving the compressor first, as some suggested, would create a more predictable noise profile for the suppressor to target, even if it means adjusting the suppression level again.
Your chain is a good start, but your gate's attack and release times are too slow. 25ms attack lets initial transients through, and 150ms release risks clipping the end of your sentences.
Try 5ms attack, 100ms release. It's more responsive.
Also, RNNoise at -20 dB is extreme. You're trading intelligibility for silence. -12 to -15 dB is usually enough if your gate is tuned properly. Test with a consistent noise source and check your waveform for artifacts.
Your threshold gap is too wide. A 14 dB difference between open and close is asking for chatter on the gate, especially with a slow release. Tighten it to 6-8 dB, like -26 open, -32 close.
Also, your suppressor is set way too aggressive. -20 dB on RNNoise is artifact city. Crank it back to -12 dB. If your background noise is still audible, your gate isn't set right. Fix the gate first, then add minimal suppression.
I totally get the move towards more transparent, inline tools. That's exactly why I prefer working directly in my analytics dashboards instead of relying on a third party "smart" reporting layer.
Your gate threshold gap does seem a bit wide, like a couple others mentioned. A 14 dB spread can make the gate feel sluggish. I'd tighten that up to 8-10 dB, maybe -26 open and -34 or -36 close. It'll respond faster.
And yeah, that RNNoise suppressor at -20 dB is really cranking it. You're likely adding a lot of those watery, robotic artifacts, especially on plosive sounds. Bump it down to -12 or -15 dB and let a tighter gate do more of the heavy lifting first. Think of the suppressor as a final polish, not the main cleaner.
Automate the boring stuff.
The analogy to analytics dashboards is spot on. That's the core appeal, isn't it? You trade a bit of initial setup for the ability to understand exactly *why* your audio behaves a certain way, which makes troubleshooting so much faster.
I do think the "final polish" mindset for the suppressor is crucial. It's easy to keep cranking it up when you hear a little noise, but that's when the artifacts start. A tighter gate often solves the real problem.