ACX Audio Submission Requirements: A Practical Preflight

Passing ACX quality control is not the same as making an audiobook sound good, but it is a measurable delivery gate. A file can contain an excellent performance and still fail because its peaks are too high, its average level is outside the required range, its noise floor is too loud, or two chapters were combined into one file.

Requirements can change. The values below reflect ACX's official audio submission requirements accessed in July 2026. Recheck that page before a real submission, especially if you reuse this checklist months later.

Build the correct file set

ACX requires one chapter or section per file. Opening and closing credits are separate files. A file must not exceed 120 minutes; split a longer chapter and use a continuation title. The official guidance says each file should include its spoken section header.

Create a manifest before mastering:

Order	File	Spoken header	Duration	Status
001	opening-credits	title, author, narrator	measured	pending
002	chapter-01	“Chapter One”	measured	pending
…	…	…	…	…
final	closing-credits	finality statement	measured	pending

Compare this manifest with the manuscript's table of contents. File systems can sort chapter-10 before chapter-2, so use zero-padded production names even if the platform later displays friendly titles.

Opening credits must state the audiobook title, author, and narrator in a way consistent with the title metadata and cover. Closing credits must clearly signal the end and restate the relevant credits. Do not let background music obscure the words.

Verify the retail sample

ACX asks for a retail sample no longer than five minutes, taken from the audiobook and free of explicit material. Its current guidance recommends the first five minutes for continuity, while allowing another selection agreed with the rights holder.

Choose a section that represents the actual listening experience. Do not create a trailer with music, claims, or narration absent from the book. Check that the sample begins cleanly, identifies enough context to make sense, and ends at a natural boundary before the limit.

Record its source chapter and exact time range in the manifest. If the chapter is revised, regenerate the sample so it does not preserve a superseded performance.

Measure peaks

ACX states that peak levels must be no higher than -3 dB. This headroom reduces clipping risk during encoding. Peak compliance should be checked after the full processing chain and final export, not on the raw recording.

A limiter can control rare peaks, but it is not permission to crush the performance. Heavy limiting creates pumping, distortion, and listener fatigue. Find clicks, plosives, or isolated transients that cause the highest peak and repair them before applying broad processing.

Measure every file. Copying settings from a representative chapter does not prove that a louder action scene complies.

Check average level and noise floor

The official ACX page requires an RMS level between -23 dB and -18 dB and a noise floor no higher than -60 dB RMS. Average level describes sustained program loudness differently from peak level; a file can pass one and fail the other.

Measure representative spoken sections and the complete file using a tool designed for ACX-style analysis. For noise floor, inspect room tone that is free of speech, breaths, edits, or intentional ambience. A silent digital gap is not evidence of the recording's noise floor.

Noise reduction should be corrective and conservative. Excessive processing can create metallic tails, watery consonants, or changing ambience between phrases. Remove mechanical hums and obvious defects at the source when possible, then listen after every processing step.

ACX recommends one to five seconds of room tone at the beginning and end of each file and says it must not exceed five seconds. Keep it consistent. Abrupt digital silence followed by a noisy room makes edits conspicuous.

Export the required encoding

Current ACX guidance calls for:

• MP3;
• 192 kbps or higher;
• constant bit rate, not variable bit rate;
• 44.1 kHz sample rate.

Keep a lossless production master before encoding. MP3 should be a delivery copy, not the file you repeatedly edit and re-export. Cascaded lossy encoding accumulates damage.

Inspect the exported MP3 rather than assuming the export dialog did what you asked. Verify sample rate, bit rate mode, duration, channel count, and audible start/end. Metadata tools can report these properties in a batch.

The official page currently restricts file names to standard US alphabetic and numeric characters. Use a conservative pattern with letters, digits, and hyphens, and keep the friendly chapter title in the platform metadata.

Separate technical checks from listening QC

An analyzer cannot detect:

• a repeated paragraph;
• the wrong character voice;
• a missing sentence;
• a mispronounced name;
• an edit that cuts a consonant;
• a chapter uploaded in the wrong order;
• inconsistent performance;
• a factually wrong spoken correction.

Run technical analysis and editorial listening as separate gates. The dialogue-preparation workflow reduces speaker mistakes, while the pronunciation guide protects recurring terms. Neither replaces a final listen.

Use headphones for fine defects, speakers for ordinary listening, and a phone speaker for intelligibility. Listen across chapter boundaries to catch changes in level, tone, or room noise that are hard to notice when files are reviewed days apart.

Automate a batch report

For every delivery file, capture:

• duration;
• sample rate;
• codec and bit rate mode;
• maximum peak;
• RMS level;
• measured noise floor;
• leading and trailing room-tone duration;
• checksum;
• source master version;
• pass, fail, or manual-review result.

Fail closed when a required measurement is missing. “The preset ran” is not a measurement. Preserve the report with the delivered files so a later revision can be compared to the accepted version.

Automation should identify exceptions, not hide them. A file just inside a threshold may deserve manual review if it differs sharply from neighboring chapters.

Run the final preflight

Before upload, confirm:

• title, author, narrator, cover, and credits agree;
• opening and closing credits are separate;
• every chapter or section has one file and a spoken header;
• no file exceeds 120 minutes;
• the retail sample is from the book and no longer than five minutes;
• each MP3 is at least 192 kbps CBR and 44.1 kHz;
• peaks, RMS, noise floor, and room tone meet the current official rules;
• names are conservative and order is explicit;
• every file passed a human editorial listen;
• the current ACX page was checked for changes.

If you render chapters in VoxForge, export and preserve the chapter masters, then apply platform-specific mastering and analysis to delivery copies. A single master can support more than one distributor, while each platform gets a fresh preflight against its own current rules.

Treat compliance as a reproducible report, not a last-minute visual check. The combination of a manifest, automated measurements, and human listening prevents most avoidable upload failures without confusing a numeric pass with a finished audiobook.