How to Give a Ghost or Spirit Character an AI Voice in Your Audiobook

Short answer: you cast the ghost from the regular voice library, then you build the spectral quality around the voice rather than inside it. In AudioProducer there is no reverb control, no whisper filter and no pitch shifter, so a spirit character is produced by which voice you assign, how much silence you leave around its lines, and what is playing underneath the scene. That constraint sounds like a limitation and mostly is not one, because the thing that actually makes a listener read a voice as dead is not the timbre.

What a ghost voice is actually doing to the listener

A dragon needs to sound big. A villain needs to sound like it enjoys something. A ghost has a stranger job: it has to sound like it does not belong to the room the other characters are standing in. That is a relational effect, and it lives in the contrast between the spectral line and the two lines on either side of it.

This is why the most common mistake is over-performing the ghost. Writers reach for the most theatrical voice in the library, add the longest pause they can find, and pile sound effects on the entrance. What comes out is a haunted house ride. The version that unsettles people is usually flatter than expected: an ordinary voice saying an ordinary sentence, placed somewhere the living cast clearly did not expect a sentence to come from.

The practical consequence is that the living cast is part of the effect. A ghost only sounds like a ghost because the people around it sound alive, present and slightly too warm by comparison. Cast the human characters first, then cast against them.

There is no reverb slider, and that changes the approach

Being direct about this saves you an hour of hunting through the editor. AudioProducer does not have reverb, echo, formant control, pitch shifting or an EQ chain. If you are looking for the button that makes a voice sound distant and hollow, it does not exist here.

What you have instead is voice selection, pause control at four levels, ambient soundscapes, background music and one-shot sound effects. Those are enough to produce a convincing spirit, but they work differently: you are not processing the voice, you are arranging the space around it.

Choosing a library voice for something that is not alive

Browse the Voices page on your home screen and preview rather than read. The library runs to well over a hundred voices with style descriptors attached, and several of those descriptors are directly useful here. Whisper is the obvious one, and it is a real descriptor in the library rather than an effect you apply. Calm is the sneakier choice, because a calm delivery on a line that should not be calm is deeply odd in a good way.

Two things to weigh while previewing. Pick a voice with less energy than your narrator, not more: a spirit that pushes harder than the person narrating the scene just sounds like a second narrator. And consider an accent or age band nobody else in the book occupies. An older voice among a young cast, or a British voice in an otherwise American town, does the out-of-place work for you without a single effect.

If nothing in the library fits, voice cloning is available on the same page, provided the voice is yours or you are authorized to use it. It is a slower path and rarely necessary for a secondary spirit.

Silence and ambience do the haunting

The pause system is the closest thing to a spectral control that AudioProducer has. You can drop an inline pause anywhere in the text, set a default paragraph pause for the whole project in Edit Project, and override that default on individual paragraphs where a transition needs more breath. The pattern that works for a spirit entrance is a longer gap before the line than after it. The gap before is the room noticing. Closing normally afterwards keeps the scene moving instead of turning every appearance into a stinger. Our post on pauses and dramatic timing goes deeper on the mechanics.

Underneath that, put an ambient soundscape rather than a sound effect. Wind, a low room tone, distant weather: something that runs under the passage and does not resolve. A looping bed under a spectral exchange is the most effective tool in this list, because it fills the space the reverb would have filled. One-shot effects are for the moment of arrival, used once, not on every line. Auto-Assign Sounds will make a decent first pass at all of this from the text, but treat its output as a draft: it tends to place effects on the literal nouns in the sentence, which is the wrong instinct for a scene whose horror is implicit. The ambient sound guide covers building those beds properly.

Casting the ghost while the living cast stays living

Run Auto-Assign Characters on the chapter and it will tag lines by speaker, including the spirit if the prose attributes its dialogue clearly. Where ghost scenes trip it up is unattributed speech, which is common here because the writer wants the reader unsure who spoke. Expect to re-tag those lines by hand, and check for a spirit whose lines got folded into the narrator.

Once the character exists it gets its own voice like any other and holds that assignment across the chapter. If the ghost belongs to a series, the character list including assigned voices can be imported from another project through the three-dot menu next to Add Character. The same approach applies to any non-human character, and we have written up the neighboring cases for villains and robot or AI characters.

Keeping the same ghost across a whole book

Spirits tend to appear in chapter three, chapter eleven and the last chapter, with months of drafting in between. The voice assignment persists inside a project, so the risk is not the voice. The risk is that your pause and ambience habits drift, and the chapter-eleven ghost arrives with a four second gap and a thunder crack because you were tired. Write the recipe down when you get it right: which voice, how long the entrance gap is, which soundscape runs underneath. Our guide to keeping a character voice consistent has the longer version, and if the whole book lives in this register, the horror novel walkthrough is the better starting point.

When the chapter renders the way you want, you download the audio file and it is yours to do with as you like. AudioProducer produces the file; we do not publish, distribute or list anything on your behalf. You can try the whole workflow on a free account with 1,200 words a month and no card, and paid plans start at $39.99 a month if the book is longer than a test scene.

Casting the rest of the book around the ghost is its own job: choosing AI voices for characters covers the method, and multi-voice character narration covers running the cast together.