How do I make an AI voice yell or sound angry?

Updated 2026-10-27·3 min read·Dialogue & audio

Why the read comes out flat

A voice model gets the words and nothing else. It doesn't know the bride's father just insulted the groom, or that the line comes after a slap. With no direction, it picks a neutral read. Sometimes that's fine. For a micro drama, where every scene turns on a strong moment, it usually isn't.

Tell it how, in brackets

In ShowMaker Studio's voice lines, a tag in brackets sets the delivery. [yells] or [angry] at the front of the line changes how it's performed. Think of it as a stage direction for the voice.

A few things we learned on Marry My Broken Son:

• Use one tag per idea. A line that needs to start quiet and end loud is usually two lines.
• The tag is read as direction, so keep it to how the line is said. Don't use it for script notes.
• Typos are spoken too. The voice says exactly what's written, so check the words as well as the tag.

Let Express write it

On a single dialogue line, Express writes the emotion tag for you and voices it. It's the quickest way to get a first read with some intent behind it, and you can change the tag if it guessed wrong.

We used it when a scene went wrong for a different reason. One take had three dialogue lines crammed into a four-second shot. The warning said so: six seconds of speech in a four-second shot. The lines came out rushed, and a delivery tag can't fix that.

The fix, in order

1. Fit the line to the shot first. If a check says the speech is longer than the shot, lengthen the shot or shorten the line. Delivery comes second.
2. Give each line room. In the rushed take, we kept the action (a champagne throw), trimmed the rest, and put each line underneath as its own dialogue line.
3. Build a shot per line. Shoot with AI on each line builds one new shot for that line, framed on whoever speaks it.
4. Set the delivery. Add a bracket tag yourself, or let Express tag it. One of ours came back as gently, which was right for the moment.
5. Press Play all. Hear every take in the scene before you render any video. A wrong read is cheap to fix as audio and expensive to fix once it's baked into a clip.

Listen before you pay

The order matters because video is where the money goes. A voice take can be redone in seconds. A rendered shot with the wrong performance in it means another attempt at the video price. So we treat Play all as a gate: if a line sounds off, it gets fixed before Render is pressed.

Each character keeps one voice across the whole episode, so the tag changes how they say a line while the voice stays theirs. Voices are covered in the voices docs, and if the mouth doesn't match the words, see AI video lip sync.

The whole episode is in the full guide. The rushed-lines fix is the Short My AI actors rushed their lines.