This is a custom input chain for easyeffects for noise reduction and a "broadcast" sound optimized for masculine voices with my mic, placement, acoustics, etc. I started with @MateusRodCosta input chain (now deprecated) based on this article from fedora magazine. Soon after I prompted LLM's to produce settings for NPR-like sound. To my surprise it worked quite well in subjective testing, that is what we have here. I also added the speech processor effect, and changed the order of the chain.
If this config breaks, check my current config file and comment if I need to update the gist.
If you know of ways to improve this configuration, please feel free to comment and make a suggestion.
NOTE: Some microphone's output mono which EasyEffects may interpret as left-only stereo, you can use Stereo Tools to fix this. Thank you to @Saiv46 for pointing this out
The config follows a logical "Broadcast" workflow: Clean -> Shape -> Control -> Protect.
We are currently running two AI noise reduction plugins back-to-back at the start of our chain. We pass the audio through RNNoise first, and then through DeepFilterNet. This provides aggressive noise isolation, ensuring the background is dead silent.
Instead of cutting the audio to total silence, we have set a Reduction of -12.0 dB and a slow Release of 250ms. This acts as a "soft gate." It pushes the background noise down gently between sentences. This feels much more natural to the listener than the audio constantly cutting out to "digital void" silence.
This sculpts the actual tonality of our masculine voice profile.
Band 0 (High Pass @ 80Hz): We are cutting the ultra-low rumble (like desk bumps) that doesn't contain useful voice information.
Band 1 (Cut @ 220Hz): We are cutting -2dB here. This removes the "mud" or "boxiness" common in male voices in small rooms, making the voice sound clearer without losing body.
Band 3 (Boost @ 3500Hz): We are adding +2dB here for "intelligibility." This helps our voice cut through background music, game sounds, etc so our audience understands every word.
Band 4 (High Shelf @ 10kHz): We are adding +2dB of "Air" which gives us the crisp, expensive, "studio condenser mic" sound.
This gives us a consistent "Radio DJ" presence.
Attack (15ms): We wait 15ms before the compressor clamps down. This lets the initial "punch" of our consonants (T, K, P) pass through, keeping our speech articulate.
Ratio (3:1): For every 3dB we go over the limit, only 1dB gets through. This smooths out moments when we get excited and loud.
Makeup Gain (+3dB): Since we are squashing the peaks, we use this knob to turn our overall volume back up, making the voice sound louder and fuller.
Because we boosted the treble in our Equalizer, our sharp "S", "T", and "Sh" sounds might get harsh. We target the 4kHz to 8kHz range to tame them and prevent ear fatigue. (The range specifically targets masculine sibilance which is lower than feminine sibilance)
This safety net stops our audio from exceeding -1.5dB. Even if we scream into the mic, the Limiter prevents digital distortion (clipping). It ensures our stream audio is loud and competitive, but never "broken."
I had some issues recording after applying this config, the final recording was only reproducing from the left speaker (in OBS I could see that the easyeffects source is double channel) I had to apply a new effect:
Stereo Tools and in the Stereo Matrix select
LR>LL (Mono left channel)