Voice-following teleprompter for macOS

Speak.
It follows.

LattePrompter listens while you talk and scrolls the script to where you are. Wander off for a moment, come back, and the next line is waiting at eye level.

Free & open source (MIT) · Speech stays on your Mac · Apple Silicon & Intel

rehearsal.md listening follow

Three moves, then you're rolling

1

Write

Scripts are markdown. Headings, bold and lists come through on the glass.

2

Prompt

One keystroke goes fullscreen. Flip on mirror mode if a beam-splitter sits between you and the lens.

3

Speak

A speech model on your Mac hears you and moves the text to match, word by word.

LattePrompter in follow mode: large script text with the current word highlighted in amber
Follow mode. The word you're on glows; the reading line stays put.

It waits for you

Classic teleprompters make you match a motor. Read too slowly and the text runs away; too fast and you're squinting at the bottom edge. LattePrompter matches you instead: your words are transcribed live and aligned against the script, so a skipped sentence or a two-minute tangent never loses your place. The moment you're back on text, the scroll picks you up mid-line.

The script library with collections and a welcome script
A small library for your scripts.
Settings showing the FastConformer English speech model marked Ready and Selected
Pick a model once. It's detected, selected, and kept on disk.

Nothing to send, nowhere to send it

Recognition runs on your Mac, using an open speech model you download once (about 460 MB). Audio never leaves the machine. There's no account and no telemetry, and the source is on GitHub if you'd rather check than trust.

The details that matter on camera

Three scroll modes

Follow your voice, run at a constant pace, or drive it yourself with the arrow keys.

Mirror mode

One toggle flips the text for beam-splitter glass.

Markdown scripts

Write anywhere and paste it in. Formatting survives.

Fonts built for distance

Atkinson Hyperlegible, Literata, OpenDyslexic or your system font, up to 120 px.

Say "nineteen eighty-four"

Numbers match their spoken form, so 1984 in the script hears you fine.

80 ms word latency

A streaming FastConformer model emits words while you're still saying the sentence.

Put it on your Mac

Version 0.1.0 · MIT licensed

The build is signed but not yet notarized, so macOS warns on first launch: right-click the app and choose Open.
On first open, LattePrompter offers the recommended speech model as a one-click download. All builds are on the releases page.