← All Guides · Noise Reduction
How to Lower Noise Floor Below -60dB for ACX (Without Robotic Audio)
Of all the technical specifications Audible and ACX enforce, Noise Floor (≤ -60.0 dBFS) is the number one reason home-recorded audiobooks fail automated validation or human Quality Assurance. Adjusting overall volume is simple, but fixing background hiss without turning your voice into a metallic, underwater robot requires understanding what ACX actually listens for.
1. What Noise Floor Actually Measures: Room Tone vs. Digital Silence
Your chapter's noise floor is the electrical and acoustic background sound present in your recording booth when you are not speaking—computer fans, air conditioning, fluorescent hum, microphone self-noise, and audio interface preamp hiss.
A common mistake among first-time narrators is using an aggressive Noise Gate or manually hitting "Silence Audio" (Ctrl+L in Audacity) between sentences. While that drops isolated gaps to -inf dB (pure digital zero), it causes two major problems:
- Human QA Rejection for "Unnatural Dead Air": Audible QA reviewers explicitly listen for smooth, consistent room tone at the head, tail, and natural pauses of each chapter. When the background abruptly drops from -55 dB room hiss to pure digital zero between every phrase, listeners perceive jarring "pumping" or clipped breaths.
- False Noise Floor Readings: Naive audio meters that average in pure digital-zero gaps falsely report -85 dB even when the real room tone under your voice sits at -52 dB. That is why acxcheck.com filters out artificial digital-zero windows (≤ -95 dBFS) and measures your true 5th-percentile room tone across 400ms intervals.
2. Hardware & Recording Environment Fixes First
Software cleanup can rescue a borderline -56 dB track, but if your raw recording has a -42 dB noise floor, no plugin can clean it transparently. Before touching EQ or denoisers, tighten your physical signal-to-noise ratio:
- Kill Constant Broadband Sources: Turn off HVAC vents, ceiling fans, mini-fridges, and noisy desktop towers inside your recording space during takes. Put your laptop outside the booth or use a fanless device.
- Improve Mic Distance (Inverse Square Law): Position your mouth 5 to 7 inches (roughly a fist's width) off-axis at a 15-degree angle behind a pop filter. Moving from 14 inches to 6 inches boosts your direct vocal signal by roughly +7 dB relative to room noise—equivalent to dropping your noise floor by 7 dB for free.
- Optimize Preamp Gain Staging: Set your audio interface gain so normal narration peaks between -12 dBFS and -6 dBFS. Setting gain too low forces you to crank volume by +20 dB in post-production, which drags up your interface's thermal preamp noise right alongside your voice.
- Watch Budget USB Microphones: Many entry-level USB mics have high self-noise around -55 dB. Using a low-noise XLR condenser or dynamic mic paired with a clean preamp immediately buys you 10–15 dB of headroom below the -60 dB ceiling.
3. Step-by-Step Software Workflow (Zero Metallic Artifacts)
Once your raw recording is clean and properly gained, apply processing in a disciplined order. Note that raising vocal loudness (RMS) to ACX's -20.5 dB target also raises your background noise by the same decibel amount—so always clean noise before or alongside RMS normalization.
Step A: High-Pass Filter at 80 Hz
The human speaking voice rarely carries meaningful vocal energy below 80 Hz. However, footsteps, distant traffic, desk vibrations, and HVAC rumble pack massive low-frequency energy that eats up peak headroom and inflates noise floor meters.
Apply a gentle 80 Hz high-pass filter (12 dB or 24 dB/octave roll-off; in FFmpeg: highpass=f=80). This alone commonly drops measured noise floor by 3 to 6 dB with zero audible change to vocal warmth.
Step B: Gentle Spectral Noise Reduction
Instead of brute-force gating, use subtle spectral denoising that analyzes room tone frequencies and attenuates hiss evenly:
- In Audacity: Highlight 2 seconds of pure room tone, run Effect → Noise Removal and Repair → Noise Reduction, click Get Noise Profile, then select the entire chapter and apply Noise Reduction: 6 dB, Sensitivity: 6.00, Frequency Smoothing (bands): 3.
- In FFmpeg / Automated Pipelines: An FFT denoiser such as
afftdn=nf=-25gently suppresses steady-state broadband hiss while leaving vocal transients intact. - Why Less Is More: Pushing reduction to 18–24 dB strips upper vocal harmonics, leaving behind "phaser" chirps and metallic ringing on sibilants ("s" and "sh" sounds). Two light passes of 4–6 dB sound dramatically more natural than one heavy 12 dB pass.
Step C: Normalize RMS & Limit Peaks
After denoising, bring your integrated RMS into the -20.5 dB sweet spot and cap peaks with a brickwall limiter at -3.3 dBFS. Finally, verify that your finished 44.1 kHz 192 kbps CBR MP3 still sits comfortably under -60.0 dB noise floor—because MP3 compression itself can introduce slight quantization noise if your track is borderline at -60.2 dB. Aiming for -63 dB to -68 dB gives you a rock-solid safety margin.
4. Common Noise Floor Pitfalls to Double-Check
- Checking Noise Floor Before Raising Gain: If your raw voice is at -29 dB RMS and noise floor reads -65 dB, boosting the track +9 dB to reach -20 dB RMS pushes your noise floor up to -56 dB (FAIL). Always measure after mastering.
- Inconsistent Room Tone Across Chapters: Recording Chapter 1 on a quiet Sunday night and Chapter 2 while rain hits the roof creates noticeable shifts in background texture.
Test your chapter's true Room Tone Noise Floor in 2 seconds
Drop your raw or mastered chapters into acxcheck.com to inspect true 5th-percentile room tone, RMS, and Peak—or run 1-click Auto-Fix to clean 80Hz rumble and broadband hiss automatically.
Check Your Audio Now →