Head and tail silence
Every track needs a short run of silence at the start and the end. Retail players cut between files with no gap, so without that padding a chapter clips the first syllable of the next one and the listener hears the narrator jump. One to five seconds satisfies Authors Republic. ACX asks for half a second to one second at the head, with the wider range applying to the tail, so about one second at the head keeps both happy.
It has to be recorded room tone from your booth, not digital silence. A stretch of absolute zero is audible as a hole, and the listener notices the room disappear and come back. Four of the five items on the rejection report that prompted this tool were missing head or tail silence, and the cause is nearly always the same: an editor trimmed tightly to the first word and never added the pad back.
Loudness, and what −23 to −18 dB RMS means
RMS is average level over time, closer to how loud the book feels than how loud its loudest moment is. Aggregators hold every book in a narrow band, −23 to −18 dB, so a listener moving from one audiobook to the next doesn’t have to reach for the volume. Master louder than the band and the book sounds harsh. Quieter, and it disappears under road noise.
This tool measures RMS the way Audacity and the ACX Check plugin measure it, with a full-scale square wave reading 0 dB. That convention matters. The other common one reads about 3 dB higher, and the whole acceptable band is only 5 dB wide, so the wrong convention puts most files on the wrong side of a boundary. It also measures the spoken part of the track only and ignores the head and tail padding, which would otherwise drag the average down and report a failure that isn’t there.
Peak level, and why −3.5 dB is safer than −3
Peak is the single loudest sample in the file. Holding it at −3 dB leaves headroom, so that later processing doesn’t push the waveform into distortion when the aggregator re-encodes your file or a listener’s phone applies volume normalisation. MP3 encoding adds a little overshoot of its own, which means a file limited to exactly −3.0 dB in your DAW often decodes a hair above it. Limit to −3.5 dB and you have margin.
Noise floor, and why −60 dB is the line
The noise floor is how loud your room is when nobody is speaking: fans, air conditioning, the fridge in the next room, traffic, the hum of your own preamp. At −60 dB or quieter it stays below what a listener notices. Above that it becomes a hiss they hear the moment they put headphones on, and it gets worse as they turn the volume up during a quiet passage.
Record about ten seconds in your booth with no speech, using the same mic, preamp and gain you used for the book, then drop it into the room tone box above. That gives an exact measurement instead of an estimate, and it lets the tool compare your room tone against the quiet moments inside your chapters. If the pauses in your takes are noisier than your room tone sample, something was running while you recorded that wasn’t running when you captured the tone. That one is hard to diagnose from a rejection notice, because the room tone itself is clean and the takes still get bounced.
Bit rate, sample rate and constant bit rate
44.1 kHz at 192 kbps or higher is the floor for quality. Constant bit rate matters because variable bit rate files confuse some retail players’ seek behaviour, so a listener scrubbing to the middle of a chapter lands somewhere else. The bit rate also has to be identical across every file in the delivery, which is hard to catch on your own: a 256 kbps chapter sitting among 192 kbps chapters looks and sounds fine. It only surfaces as a rejection.
192 kbps is also the only bit rate at which a maximum-length chapter fits the file size cap. A full 110 minutes at 192 kbps comes to about 158 MB, under the 170 MB limit. The same chapter at 256 kbps is about 211 MB, and fails on size while passing on bit rate.
Mono or stereo, but not both
Either is acceptable. Mixing them within one book is not. A listener moving from a mono chapter to a stereo one hears the voice shift position and level, which reads as a production error even when nothing else changed.
The retail sample, and the five-minute trap
The sample has to run between one and five minutes, and the boundary is where files die. Validators report duration in whole seconds, so a file landing within about half a second of 5:00 gets reported as 300 seconds and rejected for sitting at the limit. You can’t hit exactly 5:00 anyway. MP3 frames are 1152 samples each, so at 44.1 kHz no whole number of frames lands on 300.000 seconds. Aim for 4:50 and the problem goes away.
Cover image
2400 × 2400 pixels, RGB, JPG, under 5 MB, with a filename made only of letters and numbers. A cover exported from a print layout is often CMYK, which opens fine on your machine and shifts colour when the aggregator converts it. The filename rule is duller and just as fatal: spaces and symbols break the ingestion pipeline, so My Book Cover (final).jpg gets rejected where MyBookCoverfinal.jpg goes straight through.
What this tool does not check
It reads the technical properties of your files. It can’t tell you whether the opening track announces the title and author, whether the narration matches the manuscript, whether a chapter is mislabelled, or whether the performance is any good. Every aggregator reviews those by hand. What this catches is the mechanical set: the rejections that cost weeks and take ten minutes to fix.
