A voice-following teleprompter listens as you read and moves your script to keep pace with you, holding still when you pause. In cuekards the feature is called Follow my voice. It runs in a desktop browser, using a one-time 64 MB English speech model that stays on your machine. Here's how it works, and what you can do with it.
What is a voice-following teleprompter?
A voice-following teleprompter is a teleprompter that listens to you read and moves the script to match what you've said, instead of scrolling at a speed you set in advance.
And it can change how your recordings look and sound. With a timed scroll, the prompter sets the pace and you chase it. Slow down to land a point and the text runs away from you. Stop to think and it keeps going. So you end up reading at the machine's pace, not yours.
But with voice-following, you set the pace. Speed up and the script speeds up. Take a breath and nothing moves until you start talking again.
You'll also see these tools called by other names like "voice-activated" or "voice-controlled" teleprompters.
How does voice-following work?
Every voice-following teleprompter does some version of the same three things, many times a second.
- Listen. Speech recognition turns the sound from your microphone into words as you say them, including a best guess at the word you're halfway through.
- Match. The prompter takes the last handful of words you said and looks for them in the script.
- Move or hold. If enough of those words line up with the script, in the right order, the reading position moves to the last word you said. If they don't, the prompter stays where it is.
That sounds simple. The hard part is deciding where in the script to look, and when to trust a match.
It only looks near where you are
Scripts repeat themselves. "In this video", "the next step", "let's take a look" can each turn up five times in a ten-minute script. If a prompter searched the whole script for every phrase you said, it would regularly jump to the wrong copy.
So a good voice-following teleprompter only searches a short stretch of script around your current place: a little way ahead, and a little way behind. A repeated phrase three pages later is simply out of reach, so it can't pull you there.
The downside of this is that if you skip a big chunk and start reading a different part of your script, the prompter won't be able to jump to it.
It waits for you
If you pause, the prompter waits. Mess up a sentence and you can pause, then start it again — the pause tells cuekards to forget the words you just said, so it follows you back to your restart instead of running ahead.
It runs entirely on your machine
Speech recognition can run on a server or on your own computer. For a teleprompter, running it locally on your device is better in two ways.
- It keeps up. There's no round trip to a server between you saying a word and the script moving.
- It's private. Your audio never leaves your computer, and neither does a script you paste in.
The only thing you need to do is download a speech model file to your machine first. This is what allows the teleprompter to understand where you are in the script. Currently, we only support English language.
Limitations of voice-following
Different people speak in different ways. So a good voice-following teleprompter has to be clever enough to handle these situations:
| What you do | What a simple matcher does | What should happen |
|---|---|---|
| Pause mid-sentence to think | Glues the words either side of the pause into one phrase and can misplace you | Hold your place, and treat what you say next as a fresh phrase |
| Stumble over a word | Treats the garbled word as a match for something similar and lurches | Tolerate near-misses on longer words, and hold when unsure |
| Ad-lib a sentence that isn't in the script | Hunts for a match anywhere and lands somewhere random | Hold your place, show that you're off script, and pick up when you come back |
| Go back and redo a line | Jumps forward if the line's opening words appear again later | Move back to where you restarted |
| Skip a paragraph | Loses you, or leaps to a repeated phrase | Follow short skips on its own, and give you a key for big ones |
| Say a phrase that repeats later in the script | Jumps to the later copy | Only look near your current place |
| Read a number like "2024" | Never matches the numerals to what you actually said | Compare numbers as words, and let you help by writing them the way you'll say them |
| Read a script in a language the model doesn't know | Wanders, or doesn't move at all | Say so up front, and let you switch to a fixed-speed scroll |
Two practical tips work with any voice-following prompter:
- Pause briefly before you redo a line. About a second of silence makes a restart much easier to follow.
- Write numbers the way you'll say them. If you'll say "twenty twenty-six", write that rather than "2026".
How cuekards handles this
Listening happens in your browser. When you turn on Follow my voice, cuekards asks before downloading its English speech model (about 64 MB) and keeps it in your browser for next time. Recognition runs on your machine, and it gives extra weight to the longer words in your own script, so it hears your script better than speech in general.
It looks at a short stretch of script. At any moment, cuekards compares what you just said with roughly the next 30 words and a little of what's behind you. It only moves when at least a couple of your words match the script in order. Longer words can be a near-miss and still count, which helps with accents and mumbled syllables.
A pause starts a fresh phrase. About a second of silence tells cuekards you've finished a phrase. It keeps your place but stops matching against the words from before the pause. That's what lets you pause, go back to the start of a sentence, and have the prompter move back with you. If you restart without pausing it usually still follows, but a short pause is what makes it reliable.
Going off script is fine. If what you're saying doesn't match the script nearby, cuekards holds your place and shows an OFF-SCRIPT label. When you come back to a phrase it can see, it picks up from there.
You can take over at any time. A few keys cover what the matcher can't:
- Enter re-syncs. It matches what you just said against a stretch of script about four times wider in each direction, so it finds you after a big skip. It works while a take is paused, too.
- Up and Down arrows move one line back or forward.
- Home takes you back to the top of the script.
You can change any of these keys in Settings.
Your eyes stay in one place. The line you're on sits at a fixed reading position about a third of the way down the screen, and the script scrolls smoothly underneath it. Words you've already said dim, and the word you're on is highlighted. Put that reading line as close to your camera lens as you can and your eyeline barely moves.
Notes stay out of the way. Anything in [square brackets] and any line starting with # is shown on screen but never tracked, so you can leave yourself reminders like [smile] or # show the product without confusing the matcher.
The limits, up front. Follow my voice is English only for now. Cuekards runs on desktop, in Chrome, Edge, Brave or another Chromium-based browser (version 120 or newer). It doesn't work on phones yet.
The full setup, from pasting your script to saving the file, is on how it works. The quickest way to understand voice-following is to try to break it: open cuekards, paste a script, tick Follow my voice, and read the first paragraph out loud. Pause, stumble, go back a line, and watch what the prompter does. The browser version is free, and there's no account to create.
When fixed-speed scrolling is still the better choice
Voice-following isn't the right mode for every take.
- Your script isn't in English. Follow my voice only understands English, so for anything else a steady scroll is the way to go.
- The timing is fixed. If you're reading a voiceover that has to fit a 30-second cut exactly, a constant speed you've rehearsed to can be more predictable than following your own pace.
Without the speech model, cuekards scrolls at a constant speed instead. It starts at 130 words per minute and you can set anything from 40 to 400. The scroll only runs while you're recording, so it doesn't race ahead during the countdown.
Does a voice-activated teleprompter work offline?
It depends on where the speech recognition runs. If it runs on a server, it needs a connection for every word. In cuekards, recognition runs on your machine, so once the speech model is downloaded, following your voice doesn't need the internet. Loading a script from a Google Doc still does.
Is my voice sent to a server?
Not in cuekards. The speech model runs in your browser, so your audio, your video and the recognised words stay on your computer. With other tools, check whether recognition happens "in the cloud" or "on device": cloud means your audio is uploaded.
What happens if I go off script?
A good voice-following teleprompter holds your place until you come back to the script. Cuekards shows an OFF-SCRIPT label while you're ad-libbing, then picks up when you return to a nearby line. If you've jumped a long way, press Enter to re-sync.
Does voice-following work in languages other than English?
Follow my voice is English only for now. You can still record in any language; the script just scrolls at a fixed speed you set instead of following you.
Do I need to install anything?
The only download is the optional English speech model, about 64 MB, kept in your browser for future takes. Without it you can still record with a fixed-speed scroll.