How to Make a Radio Drama With AI Voices

A radio drama is not just an audio drama with an older name. The form came out of a broadcast slot, and the slot shaped everything about it: a fixed running time, an act break where the station took a commercial, an announcer who told you what you were listening to, and a listener who was almost certainly doing something else at the same time. Those constraints produced a set of craft habits that still work, and they are the reason a well-made radio play is easier to follow than a lot of modern fiction podcasts.

This guide walks through making one with AI voices in AudioProducer: how to structure the episode, how to handle the announcer, how to move between scenes with sound instead of narration, and where the real friction is. If you want the general multi-voice production workflow rather than the radio-specific form, start with how to make an audio drama with AI and come back here for the structure.

What makes it a radio drama and not just a scene with voices

Four conventions do most of the work:

• A fixed running time. Classic slots were 15, 28, or 58 minutes. Pick one and write to it. A hard length forces you to cut the scenes that do not turn the story.
• Act breaks. Two or three acts inside the episode, each ending on a small reversal. The break is a place the listener can leave and come back.
• An announcer. A voice outside the story that opens it, names it, and sometimes closes it. Not the narrator. The announcer talks to the audience; the narrator talks about the story.
• Sound as scenery. You do not describe the room. You put the listener in it with what the room sounds like, then let the dialogue do the rest.

If your script has all four, you have a radio play. If it has none of them, you have an audio drama, which is fine, but it will be a harder listen in the car.

Step 1: Structure the episode before you cast a single voice

Write the episode as a spine of scenes with a target minute count next to each one. A 28-minute episode is roughly 4,000 words of spoken content, so a six-scene structure gives you about 650 words per scene. That number is not precious, but having it stops the middle from sprawling.

Mark your act breaks explicitly in the text, because you will need them later as pause points. A useful pattern for a three-act half hour: cold open (2 min), act one (9 min), break, act two (9 min), break, act three (6 min), tag (2 min).

Step 2: Give the announcer its own character

In AudioProducer, paste your script into a blank project or import it as EPUB, then run Auto-Assign Characters. The AI tags each line by speaker: narrator, named characters, or in-world labels. It will usually hand your announcer lines to the narrator, which is the one assignment worth fixing by hand every time.

Add a character called Announcer in the Characters panel and re-tag those lines to it. Pick a voice with more presence and less warmth than your narrator, and keep it consistent across every episode of the series. The announcer is your show's signature, and listeners recognize a series by that voice before they recognize the theme music.

If your drama has a narrator as well, the two voices need to be audibly different, not just different presets. Auditioning them back to back is the fastest way to know; see auditioning AI narrator voices for how to do that without burning your whole quota on test renders.

Step 3: Transition with sound, not with words

This is the habit that separates radio drama from a recorded short story. When you move from the harbor to the inspector's office, the temptation is to write a narrator line that says so. The radio version instead fades the gulls and the rigging, drops in a short beat of silence, and comes up on a typewriter and a ticking clock. The listener does the rest.

In practice, inside the editor:

• Run Auto-Assign Sounds first. The AI reads the scene and places music beds, ambient soundscapes, and one-shot effects from the library to match what is happening. Storms get thunder and wind; a transition gets atmosphere.
• Then go through the scene boundaries by hand. A soundscape that is correct for the scene is not necessarily correct for the seam between two scenes, and the seam is where the listener either follows you or gets lost.
• Set an inline pause of roughly 1 to 1.5 seconds at each transition. Silence is the punctuation; without it the two ambiences overlap into mush.

For a deeper pass on the sound layer itself, adding sound design to an audio drama covers picking beds and effects that do not fight the dialogue.

Step 4: Build the act breaks with pauses, not edits

An act break is a longer, deliberate silence with a musical sting on either side of it. You do not need to cut the file into pieces to get one. Set a custom paragraph pause on the paragraph that ends the act, override the project default with something in the 2 to 3 second range, and put a short music cue before and after.

The project-wide default paragraph pause, set under Edit Project, governs everything else. For a radio drama, a shorter default than you would use for an audiobook works better: the form is conversational and a long default makes the dialogue feel like it is being read rather than spoken. Consecutive blank lines collapse into a single pause, so you cannot get a longer break by adding more of them; use the per-paragraph override instead.

Step 5: Keep the cast distinguishable by ear alone

Radio's oldest problem is that the audience cannot see who is talking. Two characters with similar voices in a scene with no attribution is the failure mode, and it is worse in a car than in headphones.

Three defenses, in order of how much work they are:

1. Spread the casting. Do not put two mid-range male voices in the same two-hander. Vary pitch, pace, and accent across the scene, not across the series.
2. Write attribution into the dialogue. Characters in radio plays say each other's names more than people do in life. That is not bad writing; it is the form.
3. Give each character a fixed sonic position. Consistency across episodes matters more than variety within one. If you are running a series, keep the cast list stable.

If two voices are already too close in a finished draft, what to do when two characters sound too similar walks through the fix without a full recast.

Step 6: Ship it as a series, not as a file

Radio drama is episodic by nature, and the production advantage compounds. When you start episode two, use the three-dot menu next to Add Character to import the full character list from episode one. Voices and settings carry over, so your cast does not drift halfway through the season.

Keep the intro components consistent as well: title, author, and narrator lines each with their own configurable pause. Those few seconds at the top of every episode are what make a series feel produced. If you are planning the release schedule, serializing a novel as an audio drama podcast covers the episode-by-episode side of it.

The honest limitation, up front

AudioProducer reads EPUB files or text you paste into a chapter. It does not read a PDF script, a Final Draft file, or a scanned page. If your radio play lives in screenplay software, export or copy the text out first, and expect to spend a few minutes reformatting speaker labels so Auto-Assign has something clean to work with. That step is manual, and pretending otherwise would waste your afternoon.

Everything downstream of clean text is fast. The slow part is the writing, which is as it should be.

Start with one episode

Do not plan a season. Write one 15-minute episode with a cold open, one act break, and three scenes, and get it all the way to a finished render. You will learn more about your announcer voice and your transition timing from one complete episode than from six outlines. The rest of the catalog, including every guide referenced above, is in the full guide index.

When the script is ready, start a free project and let Auto-Assign do the first pass. Then go straight to the seams between scenes, because that is where a radio drama is actually made.