Audiobook ACX Check & Master

Give it the download links to your audiobook chapter files. It measures each one against the technical specs ACX publishes for audiobook submissions, masters it, and gives you back a 192 kbps CBR 44.1 kHz MP3 for each chapter plus a before/after report with PASS or FAIL for each requirement.

What this is not: ACX also checks every audiobook by ear (plosives, mouth noise, edits, consistency, human narration). This actor measures only the published technical specs. A PASS here means the file meets those numbers as we measure them. It does not guarantee that ACX will approve your audiobook.

Not affiliated with or endorsed by ACX, Audible or Amazon. ACX and Audible are trademarks of Audible, Inc.; used only to name the specs we measure against.

What it checks

Spec source: ACX audio submission requirements, https://help.acx.com/s/article/what-are-the-acx-audio-submission-requirements (page dated Apr 15, 2026; accessed 2026-09-25).

Requirement (ACX)	How we measure it
RMS between -23 dB and -18 dB	Unweighted RMS of the whole file, in dBFS
Peaks no higher than -3 dB	Sample peak and 4x-oversampled true peak (ffmpeg ebur128); both must be at or below -3 dB. This is stricter than ACX's wording.
Noise floor no higher than -60 dB RMS	RMS of the quietest 500 ms of the file, ignoring pure digital silence. ACX doesn't define the window; 500 ms is what Audacity's ACX Check uses.
Room tone at the start and end: ACX recommends 1 to 5 s; more than 5 s is not allowed	Time before the first and after the last 100 ms of audio that is within 22 dB of the file's overall RMS. We mark under 1 s as FAIL as well as over 5 s.
192 kbps or higher CBR, 44.1 kHz MP3	Read from the output file (codec, bitrate, CBR header, sample rate)
No longer than 120 minutes	Duration
All files mono, or all stereo	Every output uses the channel setting you choose (mono by default)

What it does to the audio

1. Trims room tone longer than 1.5 s at the start or 2.5 s at the end. Where room tone is too short, it pads with a loop of the recording's own quietest half-second. It never adds digital silence.
2. If the noise floor would end up above -60 dB, it tries gentle spectral noise reduction (ffmpeg afftdn, 12 dB). It keeps the result only if the measured floor drops by at least 1 dB and nothing else breaks.
3. Sets the gain so the RMS lands at about -20 dB.
4. Limits peaks to -4 dB, or lower if the encoded file's true peak still goes over -3 dB.
5. Encodes to MP3 at 192 kbps CBR, 44.1 kHz. It then decodes that MP3 and measures it again. The report comes from that final measurement, not from the settings.

Recordings it can't fix get FAIL with a reason, not a fake pass. For example: a noise floor that stays above -60 dB even after noise reduction, no narration detected, files over 120 minutes (split them first), or unreadable files. If the pauses between phrases are pure digital silence (a noise gate or silence removal), the row says so (cause: digital-silence pauses) rather than calling it noise: the noise floor can't be measured and there is no room tone to pad with, so re-export without the gate or fill the pauses with recorded room tone. Digital-silence pauses in a file that otherwise passes are listed under warnings. When a FAIL file was still processed, the best-effort MP3 is stored so you can hear it, and the row is marked FAIL.

Input

Only submit recordings you own or have permission to process.

• audioUrls: direct download links, one chapter per file, up to 100 per run (1 GB and 120 minutes each). Any format ffmpeg reads: MP3, WAV, FLAC, M4A and others.
• Run time: measured on Apify at the default 1 GB memory, a 13-minute chapter took about 4.5 minutes to process (run of 2026-09-25). The default run timeout is 1 hour, so for a whole book raise the timeout in the run options or split it across runs.
• channels: mono (default) or stereo.
• allowDenoise: default on.

Output

• Key-value store: one MP3 per chapter, 001-<file name>-acx.mp3, and so on. OUTPUT holds the run totals.
• Dataset: one row per chapter. Each row has verdict, reasons, before and after measurements (rmsDb, samplePeakDb, truePeakDb, noiseFloorDb, headRoomToneS, tailRoomToneS, durationS), checksBefore and checks (PASS/FAIL per requirement), inputFormat and outputFormat, processing (trim, pad, gain, limiter and denoise settings actually used), and outputUrl.

Example (measured)

"The Gift of the Magi" by O. Henry, a public-domain LibriVox recording (13 min 22 s, 64 kbps stereo MP3 at 24 kHz):

	RMS	True peak	Noise floor	Head / tail room tone	Format	Verdict
Before	-24.7 dB	-1.1 dBTP (sample peak +0.5)	-59.5 dB	0.0 / 3.65 s	64 kbps, 24 kHz	FAIL
After	-20.7 dB	-3.8 dBTP	-66.9 dB (noise reduction kept)	2.45 / 2.5 s	192 kbps CBR, 44.1 kHz mono	PASS

Privacy

Your files are downloaded into the run, processed, and the results are stored in your run's storage. They stay there for your Apify plan's retention period, then Apify deletes them. Depending on your account's storage access setting, anyone with an outputUrl link may be able to download a file, so share those links with care. We never copy or keep your files.

Who built it

Built and maintained by Madrasco with AI assistance. Your audio is processed only by standard ffmpeg filters (gain, limiter, spectral noise reduction). No AI voice or generation is applied.

Support

Report problems in this actor's Issues tab. Replies are AI-assisted, and a human owner can be reached on request.