Script to Audio with AI Voices

What script to audio means here

“Script” can mean a screenplay, a stage play, a radio script or a podcast outline. AudioDrama reads one thing: plain text where each line starts with a speaker's name and a colon. If your script follows that pattern, it can be turned into audio. If it does not, it needs a few minutes of conversion, which is what this page covers.

Each speaker gets an AI voice, so you can use it as a quick table read: hear the dialogue aloud, spot lines that sound clumsy and fix them before real actors or a recording session. It uses Fish Audio's Drama 3 model through its API; AudioDrama is independent and not affiliated with Fish Audio.

Convert a screenplay into the AudioDrama format

Screenplay format puts the character name above the dialogue, adds parentheticals and scene headings, and describes action in paragraphs. AudioDrama wants everything on one line. Here is a snippet in screenplay layout:

INT. HARBOR OFFICE - NIGHT

Rain hammers the window. NORA (30s) stares at a map.

                    NORA
          (quietly)
     They moved the shipment.

                    LEO
          (O.S.)
     Moved where?

And the same moment converted:

Narrator: Interior, harbor office, night. Rain hammers the window.
Nora [quiet]: They moved the shipment.
Leo [calling from the next room]: Moved where?

The converted version is 170 bytes once speaker tags are counted, so at least 12 seconds of credit. The steps were:

1. Join each cue to its line. The character name and the dialogue go on one line, separated by a colon.
2. Turn parentheticals into directions. (quietly) becomes [quiet]. Round brackets are not directions: they would be read aloud.
3. Decide about scene headings and action. Delete them, or shorten them into one Narrator line like the first line above.
4. Drop technical cues. (O.S.), CUT TO:, camera moves and FX cues mean nothing to a voice. Replace them with a direction if the delivery matters (“calling from the next room”).

Lines that will not parse the way you expect

The parser looks at the start of each line. If it sees a short phrase followed by a colon, it treats the phrase as a speaker. Anything else goes to the Narrator. That produces some surprises in scripts that were not written for this tool:

Note: the door is already open.
Time: 3 a.m.
INT. HARBOR OFFICE - NIGHT
Rain hammers the window.

• Note: the door is already open. creates a speaker called “Note” with its own voice. Rewrite it without the colon, or give it to the Narrator as Narrator: The door is already open.
• Time: 3 a.m. becomes a speaker called “Time” for the same reason.
• INT. HARBOR OFFICE - NIGHT has no colon, so the Narrator reads it aloud, abbreviation and all.
• Names are case-sensitive: Nora and NORA are different characters.

A quick check before generating: the cast list in the app shows every detected speaker. If you see a name that is not a character, find the line that created it.

Using it as a table read

A table read is the first time dialogue is spoken by someone other than its author. Doing it with AI voices is not a replacement for actors, but it is fast, and it exposes common problems:

• Lines that are too long to say in one breath.
• Two characters who sound alike on the page and in the voices.
• Exposition that sounds unnatural when spoken.
• Scenes that run longer than you thought.

Generate a scene, listen once without reading along, and mark the lines where you lost attention. For pacing and feeling, add directions; the page on text to speech with emotions lists the ones to try.

Size, import and cost limits

• Plain text only. No PDF, FDX or Fountain import, and no automatic detection of screenplay layout.
• Scene by scene. One script is limited to 6,000 characters and 9,000 bytes per job. Generate a feature in sections and join the MP3s in an editor.
• 16 characters per script, counting the Narrator if used, and 16 built-in voices. No voice cloning.
• Voice only. No sound effects or music.
• Credits: the greater of audio seconds and script bytes divided by 15. Failed generations are refunded. See pricing.

Steps for a single scene

1. Copy one scene's dialogue out of your script.
2. Put each turn on its own line as Name: line.
3. Convert parentheticals to [directions], remove or rewrite headings and action.
4. Open the generator, paste the text, check the detected cast and choose voices.
5. Generate, listen, adjust, and download the MP3.

The format rules in detail are in the audio drama script format guide, including a copyable template.