Will Auto-Assign Characters pick up a letter on its own?

Usually not. Auto-Assign tags lines by speaker, and an inserted letter has no quotation marks and no attribution, so the body almost always lands on the narrator. That is a reasonable default guess from the source text. Add the letter writer to the character list and hand-correct those lines in the editor to point at them.

Can I make a text message sound like it is coming from a phone?

Not with an effect. There is no audio processing layer, so there is no phone filter, no EQ, and no volume control. What you can do is give each sender their own voice so the exchange reads as two people, and mark the boundary with pauses. If the screen-ness matters, put it in the source text.

Should I read the email header out loud?

Cut it unless it carries meaning. From, To, and a timestamp take several seconds of audio and tell the listener almost nothing. If the subject line is the joke or the timestamp gap is the point of the scene, keep that one piece and fold the rest into the narration around the insert.

How to Narrate Letters, Emails, and Text Messages in a Novel

Most of a novel is running prose, and a straight read handles it fine. Then chapter nine drops in a letter from the missing brother, and chapter twelve is four screens of text messages between two characters who never say each other's names. These inserted documents are where a clean multi-voice audiobook usually falls apart, because the source text stops behaving like prose and the production tools do not notice.

Here is how we would set one up in AudioProducer.ai, including the parts the app will not do for you.

Why an inserted document breaks a clean read

A line of dialogue carries its own instructions. There are quotation marks, there is usually an attribution, and the speaker is a person who exists in the scene. An inserted letter has none of that. It has an author who is not present, no quotation marks, and often no attribution at all beyond a line of italics and a date.

Run Auto-Assign Characters on a chapter like that and the letter body will almost always land on the narrator. That is the correct default guess: the AI tags lines by speaker, and nothing in the source marks those paragraphs as belonging to anyone else. The result reads as though the narrator is reciting a document, which is exactly the flat effect you were trying to avoid. Auto-Assign is a starting point and this is one of the places it needs the most correction, same as it does with dialogue-heavy scenes that drop attribution for pages at a time.

Decide what is furniture and what is content

Before you assign anything, split the insert into the part a listener needs and the part that only exists because the document is printed on a page.

Content is the body of the letter, the actual message in the email, the words each person typed. Furniture is everything that supports the layout: an email header block with From, To, Subject, and a timestamp; a fax banner; a signature line that repeats a name the reader just heard; a date stamp on every single message in a fifty-message thread. On the page these are read in a glance. Read aloud, a full email header takes about eight seconds and tells the listener almost nothing.

Keep the furniture that carries meaning. If the subject line is the joke, or if the timestamp gap is the point of the scene, that is content. Cut the rest, or fold it into the narration around the insert so it arrives as a sentence rather than as a form.

Give the letter writer a voice, or keep it with the narrator

Once the body is clean, it is a casting decision, and this is the part AudioProducer.ai is actually built for.

Assign the letter body to the character who wrote it. Add the writer to the project's character list if they are not there yet, pick a voice on the Voices page where you can browse and preview the library before committing, then hand-correct the letter's lines in the editor to point at that character. The narrator hands off, the brother's voice reads his own words, and the narrator picks the chapter back up. Our notes on choosing voices for characters apply here without change: the letter writer is a character even if they never appear in a scene.

Keeping the letter with the narrator is a real choice too, not a fallback. If the point of the scene is that your protagonist is reading the letter, and their reaction matters more than the sender's voice, the narrator should carry it. Mark the boundary with pauses instead of a voice change. You can also lean on per-line emotion to shift the narrator's delivery inside the insert while the voice stays the same.

The same fork shows up whenever a document sits inside prose, which is why diary and journal material and footnotes and endnotes get the same treatment.

Text exchanges are two speakers with no tags

A text thread is dialogue that lost its attribution in transit. Every message belongs to somebody, and the source text tells you who by indentation, by bubble color, or by nothing at all.

Treat the thread as a two-hander. Assign every message to its sender in the editor, line by line, and the exchange plays as a conversation without a single "she wrote" being added to the manuscript. Where the source has run several messages together in one paragraph, split the line first so each message can carry its own speaker.

One honest limit: there is no audio effect layer in AudioProducer.ai. You cannot apply a phone filter, a tinny EQ, or a volume change to make a text read as though it is coming off a screen. The separation you get is voice separation, and it is enough, but it is not a processed effect. If you want the screen-ness marked, mark it in the source text or with a short pause pattern before you reach for something the app does not have.

Clean the source before it goes into the project

Projects start either blank, with text pasted in chapter by chapter, or from an EPUB import. Either way, what you paste is what gets read. There is no strip-the-furniture control, so the header blocks and the repeated date stamps come out by hand, in your manuscript or in the editor after import.

Watch for characters that carry no sound. Underscores standing in for a redacted name, a row of asterisks used as a divider, and emoji inside a text exchange all pass straight through into the read. Decide what each one should become before you generate, because deciding afterwards means regenerating.

Check the seam on both sides

The insert usually sounds right on its own and wrong where it joins the chapter. Set a longer gap going in and coming out so the listener registers the handoff. Between paragraphs, the project-wide default is under Edit Project as the default paragraph pause in seconds, and you can override it on the individual paragraphs at each end of the insert. For a beat inside a line, drop an inline pause anywhere in the text. Our guide to pauses and dramatic timing covers the levels in full.

Then listen to the last sentence before the letter and the first sentence after it, back to back. If the narrator returns and it is not obvious the document ended, widen the pause rather than changing the voice. When it plays clean, export the MP3 and move on. Distribution is on you: we generate the file and hand it over, and we do not upload or list it anywhere.

You can try the whole flow on a free account with 1,200 words per month and no credit card, which is enough to run one letter scene end to end before deciding anything. Paid plans start from $39.99 per month.