Downward Expander vs. Noise Gate for Audiobook Noise Floor

A noise gate tries to hide your room by slamming a digital door in your listener's face. A downward expander simply steps into the room and politely lowers the volume.

That difference sounds small on paper. On a finished audiobook, it's the difference between a clean, pristine master and an ACX rejection you can't explain — because your meters say you passed.

I saw this exact failure mode constantly during my years in professional audiobook quality control, and it's still one of the most common issues I see in files narrators run through Greenlit today.

One file from my QC days still sticks with me. A remote narrator was tracking in a spare room with a loud desktop fan running right next to a cheap USB microphone. Knowing the hum was bad, they tried to "fix it in the mix" by slapping a heavy, aggressive noise gate across the vocal track before submitting it.

I knew something was wrong within five seconds of pressing play — well before I ever looked at a waveform:

• Jarring vacuums. Between sentences, the track dropped instantly into absolute, dead digital silence — jarring because the human brain expects some continuous ambience.
• Gated gasps. Every intake of breath got clipped into a sharp, mechanical "psst" because the threshold was too high, forcing the gate to snap open late.
• Chopped consonants. The release time was too fast, so sentences that trailed off lost their final words. "And that's how we did it" became "And that's how we did i—."

A gate only understands two states — open or closed. It can't turn your room down, only off, then back on, then off again. The human ear is far more sensitive to that on/off contrast than to a steady hum it can tune out. An expander does the opposite: it nudges the noise floor down during pauses instead of muting it completely.

My baseline template for a dry, home-studio vocal: threshold around −35 to −40 dBFS, roughly 3 to 5 dB above the narrator's ambient room hiss; ratio between 1:1.5 and 1:2, gentle enough that it never drifts into gate territory; attack of 2 to 5 ms so it doesn't soften the first consonant of the next word; release of 200 to 300 ms so phrases and breaths trail off naturally; and a range or depth cap of −10 to −15 dB so the track can never fall into true silence.

Mic distance changes the recipe. Close-miked narrators, two to four inches away, need a slightly lower threshold but a longer release — around 300 ms — to avoid chattering on mouth clicks. Narrators sitting further back pick up more room reflection, so the threshold needs to sit higher and the release should stretch past 350 ms, or the expander slices off the natural reverb tail and creates its own pumping artifact.

Here's the counterintuitive part: a noisy room is more dangerous with a gate, not less. Going from −50 dBFS of hum to absolute silence is a 50 dB jump your ear registers as violent. Going from a pristine −70 dBFS to silence is barely a ripple — but if your room is already at −70 dBFS, you don't need a gate or an expander at all. You're already well under ACX's −60 dBFS ceiling.

The practical breakdown I use on raw, unedited audiobook tracks:

• −75 dBFS or lower: leave it alone. The room tone is a feature.
• −60 to −74 dBFS: expander only, a gentle 1:1.2 ratio, just to tuck the floor down another 6 to 10 dB.
• −50 to −59 dBFS: spectral repair first, then an expander to smooth the remaining gaps.
• Above −50 dBFS: stop turning digital knobs and fix the room. No plugin saves this without destroying the voice.

ACX isn't just a script checking numbers. Every file that clears the automated sweep still goes to a human reviewer on real headphones — and a gated file can read −90 dBFS on the meter while sounding like a trainwreck in the room. I've seen rejection notes come back reading, essentially, that the files meet the automated technical metrics for peak and RMS levels, but several chapters contain a pumping or breathing effect where the background room tone abruptly cuts to silence between phrases.

ACX's real standard isn't "−60 dBFS." It's "no distracting sounds." A gate can pass the first test and fail the second every time.

The advice I see repeated constantly in narrator Facebook groups: set your gate threshold low enough and it separates the noise from your voice. It doesn't — audio physics doesn't work that way. A gate isn't a colander that filters out hiss while holding onto your voice; it's a valve. When you're silent, the door is closed and the hum disappears, so the meters look perfect. The moment you speak, the door opens all the way, and the same hum rides in underneath your voice.

That's masked noise modulation: static hitching a ride on every word, vanishing the instant you stop. The gate didn't remove the noise — it baked it permanently into your performance and hid it in the pauses, creating a high-contrast edge that pulls a listener's attention onto the technology instead of the story.

A recent case: a spare-bedroom narrator with a fan-and-HVAC hum sitting at a raw −52.5 dBFS — an automatic ACX fail. Their first attempt used a gate at a −48 dBFS threshold, a 0.1 ms attack, a 50 ms release, reducing all the way to −∞ dB. The meter read a "perfect" −90 dBFS. The file was unlistenable — chopped consonants, gasping breaths, dead silence between lines. Result: an ACX rejection.

We stripped the gate and expanded instead: threshold −47 dBFS, ratio 1:1.8, attack 3.5 ms, release 280 ms, and a hard range cap of −12 dB so the track could never fall into true silence.

You don't need premium plugins to fix this, provided you use what you already have correctly:

• Reaper: open the stock ReaGate plugin and find the Dry slider. Blending the untouched signal back into the output caps the maximum attenuation, turning a harsh gate into a functional expander.
• Audacity: the built-in Noise Reduction tool is a multi-band expander under the hood. Back the reduction slider off to a light 6 to 12 dB instead of the default heavy setting, and it tucks away room hiss without adding artifacts.
• GarageBand: the dead end. Its gate is a single threshold slider with no attack, release, or ratio control, and it's nearly impossible to keep transparent. If you're on a Mac using GarageBand, download a free external VST or AU plugin to handle expansion instead.

Three tells mean no amount of tuning will save the file:

• The voice starts sounding hollow, thin, or metallic, because the signal-to-noise ratio is under roughly 15 dB and the expander is eating into speech itself, not just the noise around it.
• The noise is dynamic — a dog, traffic, a clicking hard drive — spiking above the threshold unpredictably instead of sitting still, so the expander tap-dances on the audio instead of taming it.
• The room has heavy reverb. Bare drywall and hardwood floors bounce the voice back into the mic, and a fast release chops that natural echo into a "funhouse" effect: dry, then a burst of room tone, then a vacuum.

Plugins enhance a good recording. They don't resurrect a dead one. If you're fighting an expander for more than five minutes just to make a chapter sound human, the fix is an acoustic blanket and a re-record, not another knob.

A noise gate tries to hide your room by slamming a digital door in your listener's face. A downward expander simply steps into the room and politely lowers the volume. Run your next chapter through Greenlit's free diagnostic before you submit — it reads your noise floor in seconds, no account required.

And if you'd rather not tune thresholds and release times by hand at all: Greenlit Studio now includes a VST3 plugin for Reaper and Audacity, so the same cleanup and mastering engine that fixed the case study above runs right inside your session — and your audio never leaves your machine.