Audiobook Loudness: RMS, dBFS and the ACX Target

Audiobook loudness is measured as RMS, the average level of your speech over time, written in dBFS (decibels below digital full scale, where 0 dBFS is the ceiling and every real level is a negative number). ACX requires your RMS to sit between -23 and -18 dBFS.[1] Many narrators aim for about -20 dBFS, the middle of that window, so a file is neither strained-quiet nor fatiguing-loud.[2] That is the whole answer to "how loud should my audiobook be." The harder question, and the one this guide is really about, is why your chapters can each pass that check and still sound like they jump in volume. The short version: passing per file is not the same as sounding like one recording, and the fix is to master every chapter to the same target, not just inside the same range.

What is audiobook loudness, and how is it measured?

Two numbers describe the level of a recording, and they are easy to mix up.

RMS is the average loudness of your narration across the file. It is the number that tracks how loud the book actually feels to a listener, because our ears respond to sustained level, not to single spikes. Peak is different: it is the single loudest instant in the file, the tip of the tallest waveform. A door knock or a plosive can drive the peak high while the average stays modest.

Both are quoted in dBFS. The scale runs downward from 0 dBFS, which is the maximum a digital file can hold, so every real recording sits somewhere in the negatives. A quiet passage might read -40 dBFS; your speech average lands higher, closer to the ceiling. When someone says an audiobook is "at -20," they mean its RMS reading is -20 dBFS. RMS is the loudness noun. dBFS is the ruler it is measured against.

ACX cares about RMS for perceived volume because that is what a listener hears as "how loud is this book." Peak matters too, but for a different reason: headroom. We will place peak in a moment, then spend the rest of the guide on loudness.

What RMS level does ACX require, and what should you actually aim for?

ACX requires RMS loudness between -23 and -18 dBFS on every file you submit.[1] That is a window, not a single figure, which is deliberate. Natural narration breathes, and a small spread is fine.

In practice, many narrators aim for about -20 dBFS, the middle of the range. Audacity's own audiobook mastering guidance normalises loudness to -20 dBFS RMS for exactly this reason.[2] Landing near the centre gives you a couple of decibels of room on either side, so ordinary variation between sentences does not push a file out of spec. Worth being clear on one point: -20 dBFS is a working target that narrators and tools use, not a number ACX publishes. ACX publishes the range. The middle is where experience tends to settle.

Loudness has two neighbours in the ACX spec, and they are easy to confuse with it. Peak must stay below -3 dBFS, and the noise floor below -60 dBFS.[1] Those are not loudness, but they sit right beside it, and getting loudness right depends on respecting both. The full ACX audio requirements cover all eight specs; this guide keeps its eyes on the average.

Why is my audiobook too quiet (or too loud) for ACX?

Two failures, opposite causes, same fix.

Too quiet means your RMS sits below -23 dBFS. This is the common one for careful narrators: you recorded conservatively to avoid clipping, kept the gain low, and never brought the average up afterwards. The performance is clean, it just never reached the ACX window. A listener would have to turn the volume up to hear you comfortably.

Too loud means your RMS is above -18 dBFS. Usually this comes from heavy processing, a hot input, or compression pushed too far. It can sound dense and fatiguing over a long listen, which is the opposite of what you want across ten hours of narration.

The fix in both directions is loudness normalisation to the RMS target, not simply dragging a fader.[2] Turning everything up by a fixed amount moves the average and the noise floor together, so a too-quiet file that you crudely boost can arrive loud enough but hissing. Normalising to a loudness target sets the average deliberately, which is what the spec is actually asking for.

How do you set your audiobook to the right RMS level?

The reliable route is loudness normalisation to your RMS target, followed by a peak limiter to hold the ceiling.[2] Normalisation sets the average where you want it; the limiter catches the occasional peak that would otherwise cross -3 dBFS. Reach for loudness normalisation, not peak normalisation. Peak normalisation only looks at the single loudest sample, so it tells you nothing about how loud the book feels and can leave your average all over the place.

Order matters, and it is the step most people get wrong. Deal with your noise floor before you raise loudness, because any gain you add raises the noise floor by the same amount. A raw take sitting at -65 dBFS of background noise, boosted by 8 dB to bring the speech up, lands at a -57 dBFS floor, which fails ACX. Clean first, lift second, and the gain budget works in your favour.

If you would rather do this by hand and see the numbers as you go, you can measure RMS in Audacity with its Peak and RMS readout before and after each step. That keeps the process concrete without turning it into a full session of manual tweaking.

Why do my chapters sound different even when each one passes?

Here is the part most loudness guides skip, and it is the one that gets books sent back.

Every chapter file can sit inside -23 to -18 dBFS and still land 4 or 5 dB apart from its neighbours. One was recorded on a fresh morning, another late and tired; the mic moved half an inch; the room warmed up and the HVAC cycled. Each file passes its own spec check in isolation. Played back to back, they audibly step up and down in volume.

That gap is where rejections hide. A per-file check confirms each chapter is inside the range. It cannot tell you the chapters agree with each other. Reviewers, though, listen across the whole book, and a book that lurches in level between chapters reads as unfinished even when every file is technically in spec. ACX's own mastering guidance makes the same point about the finished product: "all chapters/sections are brought up to matching levels, which provides a smooth listening experience."[3]

So narrators adopt a tighter rule than the spec demands. Many keep every chapter within about 2 dB of the same RMS target, so the book sounds like one continuous recording rather than a stack of separate sessions. Worth saying plainly: that 2 dB figure is a widely used best practice, not a number ACX publishes. The spec gives you a 5 dB window per file; experience says use far less of it, and use the same slice of it every time.

This is exactly where mastering to a target beats passing a range. If every chapter is mastered to the same RMS figure by design, the drift never appears, and you never have to sit with a meter comparing files one against the next. For readers who arrived here after a bounce, the guide on why ACX rejects audiobooks covers the loudness and consistency angle alongside the other rejection reasons.

How do you keep every chapter within about 2 dB of the same loudness?

The principle is simple: same chain, same target, every file. The discipline is in never making an exception.

• Process every chapter through the identical mastering chain aimed at the identical RMS target. Not "inside the range." The same number.
• If you record several chapters in one sitting, master the whole session to target first, then split it into chapters. Everything from that session inherits one level.
• Keep an archived master of your settings so a chapter recorded three weeks later gets treated exactly like the first one.
• Before delivery, measure RMS in Audacity on a few chapters and confirm they read within a decibel or two of each other, not just inside the window.

The guesswork you are removing is per-file judgement. Eyeballing a meter and nudging each chapter by hand is where drift creeps in, because a human doing it forty times will not land on the same figure forty times. Setting one target and applying it identically to every file is what keeps the book coherent, and it is the by-design route: one target, every chapter, no per-file decision to get wrong.

Does silence padding change your measured RMS?

Yes, and it catches people out.

RMS is an average taken across the whole file, room tone and all, not just the spoken parts, because that is what the RMS calculation does: every sample in the selection counts toward the average. Your head and tail silence are quiet stretches, well below your speech level, so they pull the average down. The more silence in the file, the lower the RMS reads.

The effect is small but real. The same chapter measures a touch quieter with 5 seconds of tail silence than with 1.5 seconds, purely because there is more low-level room tone averaged in. If you trim or pad after you set your loudness, your final reading can drift out from where you set it.

The practical order: pad to ACX's silence window first (ACX recommends 1 to 5 seconds of room tone at both the head and the tail of each file), then measure your RMS on the padded file. Measure the file you are actually going to submit, silences included, and the number you read is the number the reviewer's tool will read.

What are the most common questions about audiobook loudness?

Your first file is free. chapterpass masters every chapter to the same RMS target, so your book sounds like one recording. No card, no account, nothing uploaded. Try chapterpass today.