How to record an audiobook, step by step

A book’s length decides everything that follows: how many sessions, how many chapter files, what a narrator would charge per finished hour. Count the narrative words — not the copyright page or the acknowledgements — and divide by a reading pace. At a typical 155 words a minute, 9,300 words make one finished hour.

The audiobook calculator opens the EPUB, PDF, DOCX or TXT itself, lists every section so front and back matter can be left out, and gives speaking minutes, finished hours and a budget. It can also export the manuscript as a narration script to read from.

The noise in the room ends up in every pause of the book, and it is the one problem that is hardest to fix afterwards. ACX asks for a noise floor of −60 dB RMS or lower. Soft furnishings, a closed door, the computer’s fan out of the room and the heating off for the take all help; hard parallel walls and air conditioning work against you.

What decides it is the distance between your voice and the room, because gain added later lifts both by the same amount. The room test measures five seconds of your room and five of your voice with your own microphone and reports that distance; about 30 dB clears it with nothing to repair.

Work close enough to the microphone that your voice sits well above the room, and at the same distance every session, so chapters recorded weeks apart still match. Switch off any noise suppression or automatic gain in the recording software: they change the sound from sentence to sentence.

The microphone test names the device the browser actually hears, its real sample rate, your level in dBFS and the room’s floor, and plays your voice back. Speech averaging between about −30 and −12 dBFS is a workable recording level.

Set the input so your loudest moments stay well clear of full scale. Loudness is set later, in mastering; a clipped peak is cut flat by the converter and nothing afterwards restores it. Record at 44.1 kHz if you can, because that is the rate ACX takes, and keep one file per chapter: ACX wants each chapter or section in a file of its own, none longer than 120 minutes.

ACX recommends between one and five seconds of room tone at the beginning and end of each file, and says room tone spacing must not exceed five seconds. Room tone is the room itself with nobody speaking, not digital silence. Keep the recorder running for a few seconds before you start reading and after you stop; that stretch is also what you copy to cover an edit.

The room tone checker measures the head and tail of a chapter against ACX and the other platforms, then tops up a short end with your own room or trims a long one.

Take out the fluffs and retakes, long pauses and mouth noises, and listen through once without looking at the waveform. Most narrators edit in a free editor such as Audacity or in their recording software; the jobs that come up around that editing have tools here:

• Remove background noise learns a steady hum or hiss from your own quiet passage and subtracts it.
• Trim audio cuts a take to exact start and end times in seconds.
• The audio joiner puts takes together in order, and can write the whole book as one M4B with a chapter marker per part.

ACX asks for RMS between −23 and −18 dB, peaks no higher than −3 dB, a noise floor of −60 dB RMS or lower, and a 192 kbps or higher constant-bitrate MP3 at 44.1 kHz. Mastering is the chain that gets a raw chapter there: filter the rumble, bring the level into the window, then limit the peaks.

Audacity ACX mastering gives that chain with its exact settings. Or skip it: the chapter checker and fixer measures the chapter, corrects what is correctable and exports the MP3, measured again from its own bytes.

Measure every chapter before you upload it. Then check them together: chapters recorded on different days drift in level and room noise, and a listener hears a chapter that is louder than the one before it even when both pass on their own. The batch checker runs the whole book, flags chapters that differ from the book’s own median, and catches mixed mono and stereo files.

Passing the numbers is not the same as being accepted: ACX’s Audio Lab FAQ says passing analysis does not guarantee approval, because people also listen for pronunciation, extraneous sounds and consistency. If a file comes back, why was my file rejected goes through each rejection reason.