Producing an audiobook in AudLit

Details is title, author, narrator and description. Settings is where two choices actually matter:

• Output specification. ACX Audiobook is the default and is what you want for Audible. The others exist for broadcast, corporate and phone work, and each one changes the sample rate, the loudness target and the compliance thresholds AudLit will hold you to at the very end. Pick it now — it is the standard your finished files get measured against.
• Your audio folder. The folder on your own machine where every take will be written. Choose somewhere backed up: this is where your recordings live.

An empty project opens on Import Your Manuscript, with two ways in:

• Browse Files, or drag and drop — .docx, .epub, .fb2, .pdf, .rtf, .txt, .htm/.html.
• Paste Text — for a script you are copying out of an email or a web page.

DOCX and EPUB import most cleanly, because they carry real heading information — the file itself knows which lines are chapter titles. PDF is the hardest case: a PDF has no idea what a chapter is, only what a line of type looks like. For PDFs there is a Use AI to find chapters option that reads the text and works the boundaries out. Turn it on for a PDF; you rarely need it for DOCX.

Read the import preview — the five minutes that save you hours

AudLit shows you what it found as an Import Preview before anything is saved. You are checking three things:

1. Did it find the right number of chapters? One enormous section means chapter detection failed. Go back and try the AI option, or a different file format — do not carry on.
2. Are the section types right? Every section is Front Matter, Part, Chapter or Back Matter. Marking the title page, copyright page and dedication as front matter now keeps your chapter numbering honest for the rest of the book.
3. Did paragraphs survive? Open a chapter and scan it. If sentences are broken mid-line, or several paragraphs have fused into a wall of text, fix it now with the tools below.

Press Save Structure when it looks right. You can re-import later if you need to — chapters you have already recorded keep their takes.

Imports are never perfect. The Tools menu holds fixes for the common damage — every one is deterministic and shows you what it will change before it changes it:

• Fix hyphens — rejoins words a line break split (pro-cess → process) while leaving real compounds like well-known alone. It checks a dictionary rather than guessing.
• Fix stranded quotes — repairs punctuation a PDF detached from its word (Don’ t → Don’t).
• Rejoin split sentences… — merges a fragment stranded as its own paragraph back onto the sentence it belongs to. It refuses when the next paragraph opens a quotation, because that is dialogue rather than damage.
• Footnotes… — finds footnote markers so you can decide what to do with them before you are reading aloud.
• Add chapter headings, Compare with original, Generate credits… — for numbering, for checking AudLit’s text against your source file, and for the ACX credits script.

Structure suggestions — the offer that makes itself

Above the section list you may find an amber Structure suggestions panel. It is not part of the Tools menu and you did not ask for it: AudLit scans every section when the page loads and offers two merges it is confident about.

• Short paragraph runs — three or more paragraphs in a row that are each under 10 words. Usually a list, an address, or a run of one-line dialogue that the importer broke apart.
• Orphaned headers — a paragraph of 8 words or fewer whose text is repeated at the start of the paragraph below it. That is a heading the importer copied into the body rather than kept as a heading.

Both hurt cleanup for the same reason. AudLit finds your script inside the audio by anchoring on the first and last words of each paragraph, and a paragraph of four words gives it very little to anchor on. Merging gives the matcher a longer run to work with.

Each suggestion names the chapter it is in, shows you the exact paragraphs and their word counts, and has its own Merge and Skip. Merge All applies whatever you have not skipped. Nothing is written until you press Save, so you can merge, read the result, and leave without saving if you do not like it.

Dismiss all hides the panel for this visit only. The suggestions are recomputed from your text every time rather than stored, so they come back on the next visit until the text changes.

Proofread in Document View

The structure editor has two modes. Row View is one row per paragraph — good for surgery. Document View renders the chapter as flowing text the way you will actually read it, which is far better for spotting a paragraph break in the wrong place. In either view you can merge paragraphs, split a section, change a section’s type, and edit text directly.

Alongside is View Original, which opens your source file next to the parsed structure and scrolls in sync, so you can check AudLit’s version against the real manuscript.

Before you record a chapter, AudLit can scan it for words the transcriber is likely to get wrong — proper names, foreign words, invented words — and offer them for review. Approving one tells the transcriber to expect it, which means fewer false flags at the Analyze step and a cleaner cleanup.

Do not add ordinary English words, and do not add bare numbers. Numbers are handled separately, and boosting a common word makes the transcriber hear it everywhere.

Open a chapter and press Record in the project tabs. Two things are always there: your script in the middle and a sidebar on the right. A third — the waveform dock along the bottom — comes and goes.

The dock appears when there is a take loaded and you are not recording, so you will not see it during a take. That is deliberate: while you are reading there is nothing to play, and the screen should be the script. Stop the recording and the dock appears with your take in it.

At the top of the sidebar is a stepper naming the five steps for the chapter you are on — Record → Analyze → Clean up → Master → Approve. The gold button under it always names your next action. If you are ever lost, press that button.

Below the stepper the sidebar is six collapsible groups: Setup, Record, Review, Fix & Edit, Master & Deliver and Takes. On a chapter you have not recorded yet you will see only three of them. Review, Fix & Edit and Master & Deliver appear once a take is loaded, because until then there is nothing for them to act on. Only one is open at a time, and it opens the right one for you as you move through the chapter. Worth knowing early: the Analyze button lives in the Review group.

The one idea underneath all of it

AudLit never overwrites your recording. Your raw take stays on disk exactly as you performed it, and every step writes a new file beside it:

They all land in the audio folder you chose when you created the project. If a step goes wrong you delete the derivative and run that step again from the raw take. You cannot lose a performance this way, and that is deliberate — it is the reason the whole workflow is shaped as it is.

1 · Audio Settings, and the input channel

Pick your microphone. If you record through an interface, also check the input channel. If your XLR is in socket 2, AudLit has to be told that — otherwise it records the empty channel 1 and you get silence while your headphones sound perfect. There is a live meter per input, so watch which one moves when you talk.

2 · Dial in your mic

One minute, once per project, and again whenever you move the mic or change rooms. It asks what kind of mic you have and where to listen (headphones, so the speakers never end up in a take; in Chrome you pick the output right there, elsewhere you pick it in your system), shows you where to put the mic, listens to five seconds of your room, then has you read one sentence. You get three plain answers (your voice, your loudest moments, your room) and exactly one thing to change. Aim for the green band on the level bar: that is the same band the Record page and the booth show, so once the setup says you are set, the meters agree. From the iPad, tap Dial in your mic on the strip above the script: the computer runs it and the iPad shows every step, including the sentence to read at your own script size.

It will tell you to move closer before it ever tells you to touch the gain, because closer raises your voice and leaves the room where it is. If a sound pops, it says to turn the mic a little to the side, not to turn anything down.

3 · Room tone

Record at least five seconds of your silent booth. AudLit uses this to fill the gaps it creates when it removes a retake, so an edit sounds like your room rather than like digital silence. It stops on its own once the level is steady. Capture it in the same session, at the same gain, as you record.

4 · Scroll speed

Read a short passage so the teleprompter learns your pace. If you use the same booth every day, steps 2 through 4 are genuinely a one-time thing per project.

5 · Reading from an iPad or iPhone

The first setup on a new book asks where you will read from. Either way, this computer does the recording: your mic and headphones stay plugged into it. Pick the iPad when the computer is noisy (a laptop fan is the usual steady hiss in a home booth) or when it has to sit outside the door. You need an iPad, iPhone or tablet with a browser, the same Wi-Fi, and mic and headphone cables long enough to reach you, or wireless ones.

On the iPad, open audlit.com/remote, sign in with the same account, and tap the book and chapter. The computer’s screen says when it has connected, and Continue on the iPad brings “Dial in your mic” to the iPad, where you read the sound check sentence. The first time an iPad opens the remote it explains itself in three points: the computer records, you scroll the script by hand, and a fumble is kept going and marked for a retake. You can read that again any time from the iPad’s audio setup. Keep the computer awake (set it not to sleep, lid open), or the iPad loses it mid-take. You can change your answer any time from the Setup group in the sidebar: the Booth remote card is the switch.

Press R or the Record button. The teleprompter scrolls, and in Follow mode it listens and follows your voice rather than running on a timer.

When you fumble a line — mark a retake

1. Stop reading.
2. Press Enter. A green Retake marked cue flashes.
3. Go back to the start of that sentence and read it again.
4. Carry on.

Do not stop the recording. Do not start a new take. Mark it and re-read. Later, Clean up finds both readings and keeps the good one. This is the single most important habit in the app, and everything downstream is built around it.

After a take — did anything clip?

When a take stops, the Take check panel says in words whether anything clipped. A clean take says so, and reminds you that a raw take sounds quieter than a finished audiobook until mastering brings it up. If something did clip, it lists each spot with a play button. Run Analyze and each spot also shows its paragraph and the word it hit; Redo with Punch-In then re-records just that sentence. You do not need to re-record the whole take out of doubt.

Recording a chapter across several sessions

Record each part as its own take, then use Assemble Take in the Takes group to join them in order before you analyze. Assemble puts a proper gap between the parts so the cleanup engine treats them correctly. Do not clean them separately.

Where you left off

You do not have to remember. Every time you stop recording, AudLit marks the place — in follow mode it skips forward past the paragraphs you actually finished, otherwise it marks where the take ended, and it clears the mark entirely if you reached the end of the chapter.

You see it twice. An amber Resume here rule across the script at that paragraph, and a Resume at ¶N pill in the top right corner that jumps you there. The pill has an X, and it disappears on its own once you have scrolled to the mark yourself — it exists to carry you forward, so once you are past it there is nothing left for it to offer.

To put one somewhere deliberately, select any text and press Bookmark on the Prep tab. On the paragraph that already holds the mark the same button reads Remove Bookmark.

Press Analyze. This transcribes your take and builds a word-by-word map of where every word sits in the audio.

That map is what cleanup, punch-in, the waveform editor and text selection all run on. If a feature looks greyed out or does nothing, it is almost always because the take has not been analyzed yet.

Afterwards the left panel reports a score — either Perfect Reading or N Errors Found, with an accuracy percentage beside it. Read it as a temperature, not a list of jobs. A chapter thick with proper nouns scores lower than one without, because the transcriber mishears names, not because you misread them.

The per-finding list you can click through and act on comes after cleanup, not here — see Proofread in Clean up below.

Press Clean up this take.

What it does: reads your script, finds every place you read something twice, keeps the good reading, removes the fumble, and evens out the pauses between paragraphs so the chapter breathes consistently. It also trims the dead air at the head and tail.

What it does not do: it never cuts inside a paragraph you read cleanly, and it does not touch a file on disk until you press Save.

When it finishes you get two tabs in the waveform dock: Original (your raw take) and Cleaned (the result). Listen to the Cleaned tab.

Proofread — the list that reads your cleaned audio back

Once cleanup finishes, a panel headed Proofread appears in the left column. AudLit transcribes the cleaned file a second time and compares it against your script paragraph by paragraph. Each row is a place where the two disagree, with a timestamp you can click to hear it.

Nothing in this panel has changed your audio, and nothing in it will. It is a list of places worth listening to.

Rows come in a few kinds: text that looks missing from the audio, text that looks added, and words that look misread — read as some other word. The missing ones are the ones to read first. They are the only thing in the app that has ever caught a chapter quietly losing a section of itself.

Expect the misread rows to be wrong often. They are right somewhere around a third of the time, and nearly every wrong one is the transcriber failing a name rather than you misreading it. They are shown anyway, for one reason: the buttons on each row.

Every row takes a verdict — Good catch, Not an error, Not sure or Changed on purpose — and answering one moves it out of the working list, so the list shortens as you go. Those answers are the only way the detector learns to tell a real misread from a name it could not hear, so a Not an error is worth as much as a Good catch.

Mark fixed is a separate, second thing: your own to-do list, because a confirmed error still has to be re-recorded. Both are reversible, and nothing is ever deleted — an answered row moves to Checked below, where you can change your mind.

Gap timing — the fastest way to make a chapter sound like you

There are sliders for sentence, paragraph, scene shift and heading gaps. For the three it can measure, AudLit also shows the natural gap it measured from your own reading, with a Use this button on any row where that figure differs from the value you are on. Scene shift is the exception: a scene break is white space on the page, not anything in the recording, so there is nothing to measure. Change a value and press Re-run Cleanup with new gaps — it reprocesses immediately, with no re-recording.

Individual gaps

Between any two paragraphs in the script view there is a thin gap line. Click it to nudge that one gap by ±100 ms or ±250 ms, or type an exact number. Useful for a dramatic beat before a chapter’s last line.

Press Save Cleaned when you are happy. That writes _cleaned.wav.

When cleanup gets it wrong — tell it so

Everything above assumes cleanup was right. Sometimes it is not: a line you read once and it removed, a fumble it left in, a pause it misjudged. There is a channel for exactly that, and it is the only way anyone finds out. The narrator who loses a sentence, shrugs, and re-records it has fixed that chapter and taught us nothing.

It appears in the right sidebar as Cleanup Issues, and only after a cleanup has finished on the chapter — a report is pinned to the run that produced it, so we can replay the exact decisions the engine made. A complaint with no run behind it is a shrug we cannot act on.

There are three ways to flag something, depending on how big it is.

• A word or a phrase. Select it in the script, then open the Report tab in the selection toolbar — the third tab, beside Prep and Edit. Two buttons: Mark missing if it is not in the cleaned audio, and Mark repeated if you can hear it twice.
• A whole paragraph. Hover it and click the small flag on its right edge — the mirror of the star on the left. Same two choices.
• A gap. Open the gap line between two paragraphs, the same popover you nudge timing with. It carries two flags: gap too long, gap too short.

Nothing is sent yet. Each one lands in the sidebar panel, which forces itself open with an amber count the moment anything is waiting — a flagged issue cannot sit there unnoticed. Click a row to jump back to it, and add a note to any of them if the position alone does not explain it. When you are done, press Report 3 issues and they go together, with one optional Overall feedback box for anything that does not belong to a single spot. Nothing flagged and something still to say? The button reads Send feedback and the box is the whole report.

Sent reports stay listed under Submitted reports in the same panel, so you can see what is still open. Delete one if you change your mind; you can always re-file.

Three tools, in the order you will use them.

A · Punch-In Re-Record, for re-performing a passage

1. In the Fix & Edit group, press Punch-In Re-Record. Or select the text in the teleprompter, open the Edit tab on the toolbar that appears, and press Re-record — that works on a word, a sentence, a range of sentences or a whole paragraph, and it works out which you meant.
2. AudLit plays you a pre-roll: the seconds of your own audio leading into that spot, so you match your energy and pace.
3. A banner counts the pre-roll down, then turns red and says Record now. Read the passage.
4. Press Preview Splice. It plays you the join.
5. If the join is slightly off, drag the four bracket handles in the visual editor — or use the nudge buttons (−50 / −10 / +10 / +50 ms) — and preview again. Nothing is committed until you accept.
6. Press Accept Splice. It re-analyzes automatically so the word map matches the new audio.

B · The waveform editor, for surgical fixes

Press the expand icon on the waveform to go full screen. Select a region and Delete, Trim or insert Silence, with undo and redo.

You can also drive it from the script: select words in the teleprompter and AudLit finds and highlights them in the waveform for you. That is usually easier than hunting visually — select “the cat sat on the mat”, see exactly where it is, delete it if you want. Press Save Edits to write _edited.wav.

C · Notes to yourself, for a pass you do later

Select text, open the Edit tab, press Add Note — and optionally record a voice memo with it (“this is pronounced KAI-ro”). It appears in the sidebar list.

Then press Queue next to Punch-In Re-Record and AudLit walks you through every open note in order: it navigates to each one, arms the punch-in, and moves on when you accept. This is the efficient way to fix a whole chapter — mark everything on one listen-through, then do them all in a single pass.

Press Master. AudLit runs the full chain against whichever output spec the project is set to.

Or press Master for ACX (automatic). It masters with your settings first. If the chapter passes, that is the result. If only the noise floor fails, it tries the next-cheapest fix, one at a time, and stops at the first that passes: a slightly quieter level (still inside ACX’s range), then noise reduction at increasing strength. The result lists every attempt and what it used. It never changes your book’s settings by itself; if it had to change something, a button offers to use those settings for the rest of the book.

At the top of the panel is a card naming exactly which file it is about to master — Edited before Cleaned before Raw. Glance at it. After several edits it is the quickest way to confirm you are mastering the right version.

You then get a compliance report. For ACX:

Then Save WAV and Save MP3. The MP3 is your ACX upload. AudLit measures the MP3 itself after encoding and shows the result under the report, so you know the file you upload passes, not just the audio it was made from.

Settings for the whole book

Settings → Mastering holds how every chapter of the book is mastered: noise reduction and its strength, loudness (standard or a little quieter), EQ, tone range, and the silence at the start and end of each chapter. They are the same settings the mastering panel uses, and they apply to batch export too. A change applies to chapters you master from then on, so re-master any you have already done if you want the book to sound even.

New books start with noise reduction on, using the Adaptive method, which learns your room’s noise from the take itself. A book started before it arrived keeps the settings it had until you change them, so nothing about a book in progress changes underneath you.

If someone else has to sign off — an author, a rights holder, a producer — they get their own read-and-listen page, and their notes arrive in the same sidebar list as the ones you leave yourself.

Inviting them

On the project’s Review tab, type their address under Invite by email and press Invite. The link is copied to your clipboard at the same time, which matters more than it sounds: invite mail lands in spam often enough that pasting the link into a message you are already writing is usually the faster route.

What they hear

The mastered MP3, and you do not have to send it anywhere. Saving the MP3 in Master & Deliver uploads a review copy at the same time — that is what the Audio uploaded for author review message means. Chapters you mastered outside AudLit have an upload button of their own, per chapter, in the same panel.

What they can do, and what they cannot

They read the script with the audio, select a line, and leave a note — typed, or spoken into their own microphone, which is usually the faster way to convey a pronunciation. They can reply on a thread. They cannot mark anything resolved: a note is closed by the person who has to act on it, which is you.

Per chapter they give a verdict, Approve or Request Changes with a description of what needs to change. Their notes land in the sidebar beside your own, and the Queue from Fix & Edit walks them the same way.

Their approval is not the same as yours

They are two separate things and it is worth keeping them straight. The author approving a chapter records their sign-off; it does not lock anything. Locking is the next step, and it is yours. When several chapters have come back approved, Apply approvals in the review header turns the whole batch into locked chapters in one press — it appears only for you, and only when there is something to apply.

What gets emailed, and when

Worth knowing before you invite anyone, because it surprises people: once an author has accepted the invite, every time you save a mastered MP3 they get a mail naming that chapter. That is the point of the feature, but it means a chapter you re-master four times sends four messages. Notify Reviewers is the other direction — one mail covering everything that is ready, useful when you have uploaded a batch by hand.

The notification setting in project settings is narrower than it looks. It governs the approval mail only: on immediate, each chapter an author approves sends one; on manual, those are held for a summary you send from the Overview page. It does not quiet the new-audio mail. And until somebody has accepted an invite, none of this sends anything at all — a pending invite is not a recipient.

Approving marks the chapter final and locks it. Nothing is emailed to anyone unless you have reviewers set up on the project — and the panel tells you which case you are in before you press it. You can reopen an approved chapter from the Master & Deliver group if you need to.

1 · A bad import costs more than a bad take

Everything downstream compares your audio against the text AudLit holds. If a chapter imported as one giant paragraph, or two chapters fused into one, cleanup has the wrong map and will behave oddly in ways that look like an audio problem. Spend five minutes on the import preview. It is the cheapest time you will ever spend.

2 · Work from the raw take

If a chapter sounds wrong, do not clean the cleaned file — go back to _take01.wav and clean that again. AudLit handles this for you automatically now, but it is the right mental model, and it is why the raw take is never deleted.

3 · Room tone is a snapshot, not a fact about your room

If the air conditioning starts halfway through a session, the room tone you captured cold no longer matches what you are recording. Run Dial in your mic at the start of a session rather than trusting last week’s capture: it listens to the room again and offers the fresh room tone. Mastering is more forgiving here: Adaptive noise reduction starts from your room tone but then follows the noise in the take itself.