ACX wants room tone, not silence. Its current requirements recommend 1 to 5 seconds of room tone at the beginning and end of each file, and room tone spacing anywhere in the file must not exceed 5 seconds. So when you remove silence from narration, you shorten pauses and replace edits with clean room tone. You never cut a gap down to digital zero. This post covers the ACX numbers, how to handle pauses and breaths, and how to check a file before you upload it.
What ACX asks for
These are the audio requirements from the ACX help center page. Read the page itself before you submit, since ACX updates it.
| Requirement | ACX wording, summarized |
|---|---|
| Room tone at head and tail | 1 to 5 seconds recommended at the beginning and end of each file |
| Spacing | Room tone spacing must not exceed 5 seconds |
| Loudness | Between -23 dB and -18 dB RMS |
| Peaks | No higher than -3 dB |
| Noise floor | No higher than -60 dB RMS |
| Format | 192 kbps or higher CBR MP3, 44.1 kHz |
| Channels | All files in the same format, all mono or all stereo |
| Length | No file longer than 120 minutes |
| Content | One chapter or section per file, with its section header; opening and closing credits as separate files |
| Clean audio | No plosives, mic pops, mouse clicks, excessive mouth noise or outtakes |
Note on the head and tail figure: an older ACX blog post on editing gave 0.5 to 1 second at the head and 1 to 5 seconds at the tail, and an editor’s note on the same post points to a newer rule. The requirements page is the current reference, and it calls its 1 to 5 second figure a recommendation.
Room tone, not digital silence
ACX’s glossary defines room tone as the background noise in a room, and says that for audiobooks it should be the resting sound in your studio, as close to silence as possible. That last part is easy to misread. As close to silence as possible means a quiet room. It does not mean replacing the room with zeros.
ACX’s own QA post is direct about this: “Your audio should never drop to absolute zero.” It adds that your room tone should be your baseline and must stay consistent throughout. The reason is simple. Listeners hear the room under your voice the whole time. When a gap drops to digital silence, the room vanishes and comes back, and that switch is more noticeable than the room ever was.
ACX’s post on editing and spacing says the same from the editing side: portions you edit out are replaced with clean room tone. It also advises recording new room tone whenever you change your microphone or studio settings.
How to record room tone
- Set up exactly as you will for narration: same mic position, gain and room.
- Stay still and quiet, and record a stretch of the empty room.
- Label it with the session date, and record it again whenever anything in the setup changes.
- Use it to fill every edit and gap in files from that session.
Which pauses to shorten and which to keep
Pauses in narration are part of the performance. They tell the listener where a sentence ends, where a paragraph turns and where a scene changes. Silence removal should take out the hesitations and dead spots, not the structure.
| Pause | What to do |
|---|---|
| Between words, from a stumble or hesitation | Shorten or remove |
| Between sentences | Keep a natural pause, shorten only obvious drags |
| Between paragraphs | Keep it noticeably longer than a sentence pause |
| Scene breaks and section changes | Keep the longest internal pauses here, under ACX’s 5 second spacing limit |
| Head and tail of each file | Room tone, following the current ACX recommendation |
| Retakes and outtakes | Remove the bad take and its pause, fill the joint with room tone |
ACX does not publish fixed lengths for sentence or paragraph pauses. The best guide is consistency: decide on a feel for each kind of pause in chapter 1 and keep it through the whole book. ACX lists consistency of spacing among its requirements alongside levels, tone and noise.
Breaths
Breaths are where narration editing goes wrong most often. Remove them all and the voice sounds like it never needs air. Leave them all and a quiet passage fills with inhales.
- Remove loud or sharp breaths, and the breath before the first word of a file.
- Lower the volume of the remaining breaths instead of deleting them, if your editor lets you.
- When you delete a breath, fill the space with room tone, not silence, and keep the pause length about the same.
- Watch for breaths that a silence remover keeps because they are louder than its threshold. They often sit right at the start of a sentence.
What automatic silence removal does to narration
Most silence removers use three settings: a level below which audio counts as silence, a minimum pause length, and padding kept around speech. Applied to narration with podcast-style settings, they produce pauses that are all the same short length. That flattens the difference between a comma, a full stop and a paragraph break.
Two approaches work better for long-form narration:
- Shorten, don’t remove. In Audacity, Effect > Special > Truncate Silence has two modes. Truncate Detected Silence cuts every detected pause to the same length. Compress Excess Silence shortens only the part of each pause beyond the minimum duration, by a percentage you set, so longer pauses stay longer than short ones. The second mode keeps the shape of your pacing. Because both modes shorten the existing pause, the audio that remains is your own room tone. The Audacity guide has the full steps.
- Remove, then rebuild the head and tail. The free tool on this site removes pauses longer than its Shortest pause setting and keeps a margin set by Keep around speech. It also trims silence at the very start and end of the file completely, so you need to add room tone back to the head and tail before submitting.
Set the silence level a little above your noise floor. If the level is below the room, nothing counts as silence. If it is far above, quiet word endings get cut. Silence threshold settings explains how to find the right level.
Loudness: ACX measures RMS, not LUFS
ACX’s loudness rule is -23 to -18 dB RMS with peaks no higher than -3 dB. The requirements page does not mention LUFS. That matters because many tools normalize to a LUFS target, including Audacity by default.
QuellCut, for example, normalizes audio to -16 LUFS. LUFS is a different measurement from RMS, and -16 LUFS is not the ACX spec. A file normalized to -16 LUFS is not set to the ACX range, and nothing about the normalization checks your peaks or noise floor. If your book goes to ACX, measure RMS and peaks after any normalization, and adjust with a tool that targets RMS.
Audacity’s Effect > Volume and Compression > Loudness Normalization can do that. It offers Perceived loudness (default -23 LUFS) or RMS (default -20 dB). RMS normalization raises or lowers the whole file, so check peaks afterward. For more on LUFS targets, see Podcast loudness and LUFS.
A workflow from recording to upload
| Stage | What to do | How to check |
|---|---|---|
| Record | Same setup every session; record room tone at the start of each | Listen to the room tone on headphones for hum and noise |
| Edit | Remove retakes and stumbles, fill every joint with room tone, shorten long hesitations | Listen at 1x to the edited sections, not only at speed |
| Pauses | Shorten pauses with a mode that keeps their relative length | Spot-check paragraph and scene breaks, confirm none exceed 5 seconds |
| Master | Bring RMS into -23 to -18 dB and keep peaks at or below -3 dB | Measure RMS, peak and noise floor on the finished file |
| Export | 192 kbps or higher CBR MP3 at 44.1 kHz, same channel format for every file | Check file properties and length under 120 minutes |
| Upload | Run the ACX audio analysis tool on each file | Read its report, then listen for what it cannot catch |
Two tools help at the check stage:
- ACX Check is an analysis plugin for Audacity, available from Audacity’s plugin site. It reports peak level, RMS level and noise floor, which it measures as the RMS level of the quietest 500 milliseconds in the selection. Its plugin page notes that this reading can be misleading when one part of the selection is quieter than the rest.
- ACX’s audio analysis tool runs on files you upload to ACX. According to ACX’s FAQ, it checks RMS, peak levels, bitrate, bitrate method, sample rate, mixed channels, duplicate files and spacing. It does not detect noise floor issues, editing problems or extraneous noise, and its results are visible to producers and DIY rights holders only. ACX also notes that RMS and peak readings can differ between DAWs and meters.
Neither tool listens to your performance. A file can pass every number and still have a clumsy edit in chapter 12.
Commercial voiceover: ads and e-learning
Outside audiobooks there is no single spec, so the client’s delivery notes come first. A few habits still apply:
- Ads are cut to a fixed slot. Shorten pauses to fit the time before you consider speeding up the read, which changes the voice.
- E-learning often runs under on-screen text and animation. Keep pauses after key points, since the learner may still be reading. Ask whether the client wants files split by slide or screen.
- Explainers and app videos are timed to picture. Pauses that look long in the waveform may be needed to hit a visual change.
- Delivery. Unless the client asks otherwise, leave a short margin of room tone at the start and end of each file so their editor has a handle to trim, and never deliver with digital silence at the edges.
Mistakes to avoid
- Muting gaps instead of filling them with room tone.
- Using room tone from a different day or setup, so the background changes mid-chapter.
- Running podcast-style silence removal on a whole book and flattening every pause to one length.
- Taking a normalizer’s LUFS result as proof of ACX loudness.
- Cutting the head and tail to zero, then forgetting to add room tone back.
- Mixing mono and stereo files in one project.
Where the tools on this site fit
The free tool on this site is useful for a first look: open a chapter excerpt of up to 3 minutes, see how much of it is pause, and hear what different Shortest pause and Keep around speech settings do. It runs in your browser and exports WAV. It does not add room tone or check ACX specs.
QuellCut on iPhone and iPad suits voiceover drafts and spoken-word content: it removes silence and filler words, lets you review every cut on a timeline, and exports M4A or WAV. Its -16 LUFS normalization is not the ACX loudness spec, and it does not make a file ACX compliant. For an audiobook, finish the master in a DAW and check it against the ACX requirements yourself. More on audio-only workflows is in the audio guide.
Questions
- How much room tone does ACX want at the start and end of a file?
- The current ACX submission requirements recommend between 1 and 5 seconds of room tone at the beginning and end of each file, and say room tone spacing must not exceed 5 seconds. Older ACX blog posts give different head and tail figures, so go by the requirements page.
- Can I use digital silence between paragraphs?
- No. ACX says your audio should never drop to absolute zero and that room tone should be your baseline. Fill edits and gaps with clean room tone recorded in the same session.
- Does normalizing to -16 LUFS meet ACX loudness?
- Not by itself. ACX specifies -23 to -18 dB RMS with peaks no higher than -3 dB, measured as RMS, not LUFS. Measure RMS and peaks after any normalization.
- Does the ACX audio analysis tool check noise floor?
- According to ACX's FAQ, no. It checks RMS, peak levels, bitrate, bitrate method, sample rate, mixed channels, duplicate files and spacing, so check the noise floor yourself.
- Should I remove every breath from narration?
- Usually not. Remove loud or distracting breaths, and reduce others rather than deleting them, so sentences do not run into each other.
Sources: ACX: Audio submission requirements, ACX: Audio analysis tool FAQs, ACX blog: Why did my title fail QA? Part 5, ACX blog: Editing and spacing with Alex the Audio Scientist, Audacity Manual: Truncate Silence, Audacity Manual: Loudness Normalization.