How to make a ghost-story audiobook with AI voices

Short answer: convert or paste your manuscript, let Auto-Assign Characters tag the speakers, cast the dead character from the ordinary voice library rather than reaching for something processed, and build the wrongness out of silence and the sound of the room instead. The harder part is the decision underneath that, which a ghost story forces on you in audio and never forced on you on the page.

Why a ghost story is a different audio problem than horror

Horror wants the listener certain that something is in the house. A ghost story wants them unsure, and it wants them unsure for three hundred pages. That is the whole engine of the genre: the governess who may be imagining the children's visitors, the widower who hears his wife in the next room at the hour she used to come home. The moment a listener stops wondering, the book has spent its only real asset.

This has a counterintuitive production consequence. A ghost story generally wants fewer effects than a horror novel, not more. Horror can afford a loud mix because it is not trying to keep a question open; it is trying to land an answer hard. Every extra layer you put under a ghost story is another vote that something is definitely there, and the listener counts those votes whether you meant them to or not. If your book actually is horror and wants the heavier sound design, our horror audiobook guide and our cosmic horror guide are the better starting points, and the full guide index has the rest of the atmospheric genres.

The decision audio forces that the page never did

On the page, a line of ghost dialogue is quoted text. Attribution is optional, and a good ghost-story writer leaves it off on purpose. The reader gets a sentence in quotation marks with nothing telling them who said it, and the doubt survives intact.

Audio has no equivalent of that. Every line in the finished file comes out of some voice, and assigning a voice is an assertion that a specific someone spoke. There is no unattributed setting. This is the single production decision that separates this genre from every other one, and it is yours to make consciously rather than by default.

In practice it comes down to a per-scene choice on the lines the prose deliberately left open. You can leave the line with the narrator, which keeps it inside their head and preserves the doubt, or you can give it the ghost's own voice, which settles the question for the listener in that moment. Both are legitimate; what is not legitimate is letting Auto-Assign Characters make the call for you and finding out on the listen-through. Auto-Assign is very good at tagging clearly attributed dialogue and it will hand unattributed speech to whoever it can infer, which in a ghost story is exactly the sentence you most wanted to control.

So make a short list per chapter: which open lines stay with the narrator, which go to the ghost, and roughly where in the book the balance is allowed to shift. Most ghost stories tip slowly from the first column to the second, and knowing where your tipping point is turns a vague atmosphere problem into a casting sheet. Gothic novels lean on the same unreliable-narrator machinery for a whole book, and our gothic novel guide covers the confined-cast version of it.

Cast the narrator first, because the narrator is most of the book

Ghost stories are overwhelmingly first-person and interior. The dead character may have three dozen lines in the entire manuscript. The narrator has everything else: the description, the doubt, the rationalizing, the part where they explain to themselves why the door was open. In runtime terms the narrator is the book, and the ghost is a handful of minutes.

That inverts the casting order most people use. The instinct is to spend the audition time on the ghost, because the ghost is the interesting problem. The listener's actual experience is decided by the voice they hear for hours. Audition narrators on the Voices page against a real passage, and pick a passage where the narrator is describing something they are not sure they saw, rather than a neutral paragraph. A voice that sounds fine reading description can sound falsely confident the moment it has to be uncertain.

Look for a narrator with slightly less push than you would pick for a thriller. Ghost stories are told by people trying to sound reasonable, and a narrator who performs the fear does the work the prose was supposed to do. Once the narrator is set, cast the ghost against it rather than in isolation. Casting the spectral voice itself has its own set of rules, and our post on giving a ghost or spirit character a voice is the detailed version of that job.

Let the room do the haunting

AudioProducer has no reverb, no pitch shifting and no EQ chain, which for this genre is closer to a feature than a limitation. What you have is voice selection, pause control at four levels, ambient soundscapes, background music and one-shot sound effects, and a ghost story only needs the middle two.

The pattern that works is one continuous low bed that never resolves, running under the scene rather than arriving with the ghost. Weather against a window, a house settling, traffic two streets away at the wrong hour. It should be the kind of sound the narrator would not remark on, because the second the bed becomes noticeable it becomes evidence. Save one-shot effects for the rare moment you genuinely want the question closed, and use a longer pause before an open line than after it.

Auto-Assign Sounds gives you a first pass, and it tends to place effects on the literal nouns in the sentence. In a genre whose scares are implicit that instinct is usually wrong, so treat its output as a draft and pull most of it back out.

Producing a ghost story chapter by chapter

Convert or paste your manuscript, work one chapter at a time, and listen all the way through before you move on. The listen-through matters more here than in plainer genres, and it is looking for something specific: the places where the mix answered a question you wanted left open. Those are almost never in the scary scenes. They turn up in the quiet ones, where a stray effect or a confidently-cast line closed a door the prose had left ajar.

If the book is part of a series, lock the narrator and the ghost early and write down the recipe while it is working: which voices, how long the entrance gaps are, which bed runs underneath. The character list with its assigned voices can be imported into a new project, so book three sounds like book one without you remembering anything. Because cost scales with the words you narrate rather than with studio hours, adding a volume to an established series stays predictable.

What you export and where it goes

When a chapter renders the way you want, you download it as a standard MP3 and it is yours. We produce the audio; we do not publish, distribute or list anything on your behalf, and you keep the copyright to your text and to the audio we help you make. One honest note on distribution: ACX and Audible currently require human narration as a sourcing rule, so an AI-narrated audiobook is not a fit there. Plenty of authors sell directly from their own site or through platforms that accept AI narration, but policies change, so check each platform's current stance yourself. This is general information rather than legal advice. On voice cloning, we only support cloning your own voice or one you have explicit permission to use.

The cheapest way to find out whether your ghost survives being read aloud is to put one scene through it. Pick the chapter where the listener is supposed to be least sure, not the one with the best scare. Create a free account and run it on the free tier, which gives you 1,200 words a month with no card; paid plans start at $39.99 a month when you are ready for the whole book.

Frequently asked questions