How to Pass the ACX Audio Quality Check (and Fix Failures)

What the ACX Audio Check Actually Measures

When you upload your audiobook to ACX, an automated analysis runs on every file before a human ever listens to it. If any file fails any single measurement, the whole submission bounces back. There's no partial credit.

Here's the complete list of what the automated check verifies:

The frustrating part: ACX's rejection emails often name the failed spec but not where in the file it failed or by how much. A file can fail RMS by 0.3 dB and you'll get the same generic message as a file that's 10 dB off.

That's why it pays to measure your files yourself before you upload. You can drop a file into our free ACX Audio Checker and see your exact RMS, peak, and noise floor numbers against every spec in the table above — in your browser, before ACX ever sees the file. It takes about ten seconds per file and eliminates the multi-day resubmission loop entirely.

Now let's go through each failure, why it happens, and how to fix it.

Failure 1: RMS Level Out of Range

The spec: RMS between -23 dB and -18 dB. This is the single most common failure.

RMS measures the average energy of your file — roughly, how loud it feels over time. ACX wants every file in a 5 dB window so listeners don't reach for the volume knob between chapters.

Why it fails:

• You normalized peaks (to -3 dB or 0 dB) but never measured average loudness. Peak normalization and RMS are different things — a whispery performance peak-normalized to -3 dB can still sit at -30 dB RMS.
• Your mic gain was set low and you never compensated in post.
• Different chapters were recorded on different days at different distances from the mic, so some pass and some don't.

The fix: You need RMS normalization (sometimes labeled "loudness normalization"), not peak normalization.

1. Measure your current RMS. If you're at -27 dB RMS, you need +7 dB of gain to hit -20 dB (the safe center of the range — aim here, not at the edges).
2. Apply that gain. But raising a quiet file by 7 dB also raises its peaks by 7 dB, which usually pushes them past -3 dB.
3. So follow the gain with a limiter set to a -3.5 dB ceiling (the extra 0.5 dB is safety margin for MP3 encoding, which can nudge peaks upward).

The gain-then-limit combination is the standard ACX mastering chain. Gain alone breaks your peaks; limiting alone doesn't raise your RMS.

If you don't want to build that chain in a DAW, the ACX preset in our converter applies RMS normalization to -20 dB and a peak limiter in one pass, along with the format conversion — and because it runs entirely in your browser, your unreleased audiobook never uploads anywhere.

Failure 2: Peaks Above -3 dB

The spec: No sample in the file may exceed -3 dB.

Why it fails:

• Plosives — p and b sounds hitting the mic capsule create single spikes 6-10 dB above your average peaks.
• Peak normalization to 0 dB or -1 dB — a habit carried over from music production. ACX wants -3 dB, full stop.
• Mono downmixing — summing two stereo channels can add up to 6 dB where the channels are correlated, creating new peaks that didn't exist in the stereo file.
• MP3 encoding overshoot — encoding can push peaks 0.1-0.5 dB above the WAV source. A file limited at exactly -3.0 dB can fail after encoding.

The fix: Run a limiter with the ceiling at -3.5 dB, and always limit after any gain change or mono conversion, and before MP3 encoding. For isolated plosive spikes, you can also manually attenuate the offending syllable by 4-6 dB — often less audible than limiting the whole file harder.

Failure 3: Noise Floor Above -60 dB RMS

The spec: The background noise during silence must measure -60 dB RMS or quieter. This is the failure people fear most, because sometimes the honest fix is re-recording.

Why it fails:

• HVAC, refrigerator hum, computer fans, or traffic in the recording room
• High-self-noise microphones (budget condensers) with the gain cranked
• Here's the trap: raising RMS raises your noise floor by the same amount. A recording with a -64 dB noise floor that needed +7 dB of gain to pass RMS now has a -57 dB noise floor — and fails.

That last point explains a very common pattern: your raw file passes the noise check, you fix your RMS failure, resubmit, and now you fail on noise. The two specs interact. This is why ACX's recommendation is to record with your noise floor closer to -70 dB, so it survives the gain you'll add in mastering.

The fix, in order of preference:

1. Kill the noise at the source. Turn off the HVAC while recording, move the computer out of the room, add soft furnishings. One re-recorded session beats hours of repair.
2. Gentle broadband noise reduction — 6 to 10 dB of reduction maximum. Beyond that you get the underwater "artifact" sound, which fails ACX's human review even when the numbers pass.
3. A high-pass filter around 70-80 Hz removes rumble and electrical hum fundamentals cheaply and transparently.

Check a silent section after your full mastering chain, not before — the post-gain number is what ACX measures.

Failure 4: Wrong Format, Bitrate, or Sample Rate

The spec: MP3, 192 kbps or higher, constant bit rate, 44.1 kHz.

Why it fails:

• Exporting VBR because it's your MP3 encoder's default (LAME defaults to VBR in many apps)
• 48 kHz files, common if you recorded in video software or on a field recorder
• Uploading WAV files directly — ACX wants MP3
• 128 kbps exports left over from an old default

The fix: This is the easiest category — it's purely a conversion setting, no artistic judgment involved. Export at 192 kbps CBR, 44.1 kHz, mono. If your recording is 48 kHz, use a real sample-rate converter (any modern tool does this properly), not a "change speed" function — the latter shifts your pitch.

Failure 5: Missing or Wrong Room Tone

The spec: 0.5 to 1 second of room tone at the head of each file, 1 to 5 seconds at the tail.

Room tone means the natural sound of your recording space with nobody speaking — not digital silence. If you gate or trim your files to absolute zero, the check can flag them, and the abrupt cut sounds jarring on playback anyway.

Why it fails:

• Trimming files tight to the first and last word
• Noise gates that clamp to digital black between phrases
• Pasting in pure silence instead of actual room tone

The fix: Record 30 seconds of room tone at every session and keep it. Paste 0.75 seconds at the head and 2-3 seconds at the tail of each chapter file. If you forgot to record room tone, copy a clean pause from within the same session's recording.

Failure 6: File Too Long

The spec: No file over 120 minutes.

This one mostly hits authors who narrate long chapters or try to upload the whole book as one file. Split at a natural pause, and remember each new file needs its own head and tail room tone. Most chapters land at 20-40 minutes, so this is rare — but a 2.5-hour omnibus "Part 1" file will bounce.

Failure 7: Missing Credits Files

ACX requires a separate opening credits file ("[Title], written by [Author], narrated by [Narrator]") and a closing credits file ("The End"-style closer plus production credit). These follow all the same audio specs as chapter files. Forgetting them, or baking them into chapter one, gets flagged during review.

A Workflow That Passes on the First Try

Order matters here. Do these steps in sequence and each spec stays fixed once you've hit it:

1. Edit your chapters in WAV or FLAC. Never do production work in MP3 — every MP3 re-encode degrades quality.
2. Convert to mono (if you recorded stereo) before any level work, so downmix summing can't create new peaks later.
3. Clean up noise if needed: high-pass filter at 75 Hz, gentle noise reduction only if the floor demands it.
4. Normalize RMS to -20 dB — the center of the range, leaving margin in both directions.
5. Limit peaks to -3.5 dB, after the gain, so nothing pokes past -3 dB post-encoding.
6. Add room tone: 0.75 s head, 2-3 s tail.
7. Export MP3: 192 kbps CBR, 44.1 kHz.
8. Verify every file — measure RMS, peak, and noise floor on the final MP3s, not the WAVs. The MP3 is what ACX judges.

For step 8, the free ACX checker reads your final files and shows pass/fail against each spec with the actual measured numbers. For steps 2, 4, 5, and 7, the ACX preset on the converter collapses them into one drag-and-drop pass. Both run entirely on your device — nothing uploads, which matters when the book isn't published yet.

If You've Already Been Rejected

Don't guess. Measure the exact files you submitted, find which spec failed and by how much, and fix only that. The most common resubmission mistake is re-mastering everything from scratch and introducing a new failure (usually the RMS-fix-raises-noise-floor trap from Failure 3).

And keep a copy of your mastering settings. When ACX requests a pickup or you record book two, running the identical chain means your files match — and pass — the same way every time.