yes, it exists. the page moves when you say the words, not when a timer says so. open prompter calls it voice tracking, it's free, and it stops the scroll after about 1.5 seconds of silence so you can go back and re-read a line.
but "follows my voice" gets used for three different things, and only one of them is what you actually want. here's how to tell them apart.
the three things people mean#
when someone asks for a teleprompter that follows their voice, they're usually describing one of these.
- timed auto-scroll. you pick a speed in pixels per second and the page moves at that speed forever. this follows a clock, not you. it's what most apps ship.
- voice-activated scrolling. the app detects that sound is happening and moves while you're loud, stops while you're quiet. it follows your volume, not your words. it will happily scroll through a sentence you never said, as long as you keep making noise.
- word-level following. the app recognises what you actually said, finds that phrase in the script, and moves the page to put it where your eyes are. this is the one that works.
the difference shows up the first time you pause mid-sentence to think. timed scroll keeps going. voice-activated scroll stops, then restarts from wherever it happened to be. word-level following knows where you stopped, because it knows what you said.
why this is harder than it sounds#
the reason most apps ship a speed slider is that word-level following has an ugly middle problem.
you do not read your script. you read a version of it. you drop a "the". you add a "so". you say "video" where you wrote "clip". you repeat a phrase because the first take of the sentence was flat. you trail off and restart. real delivery is a fuzzy, improvised cover version of the text on screen.
a system that demanded an exact transcript match would lock up on the first filler word. so the matching has to be loose enough to survive all of that, while still being tight enough to know the difference between the third paragraph and the seventh when you use the same phrase in both.
that's the whole engineering problem, and it's why the feature is rare, and why the ones that exist have a feel you either like or don't. the under-the-hood explainer goes through the loop in detail.
the 5 things to check#
if you're evaluating any app that claims to follow your voice, these are the questions that separate the real ones from the volume detectors.
1. does it match words, or just detect sound? hum a tune with your mouth closed. if the page scrolls, it's listening for volume, not language. that app will lose your place the moment you improvise.
2. does the recognition run on the device or in a cloud? this is a privacy question and a reliability question at once. a cloud recogniser needs a connection, adds latency, and means your script and your voice leave the phone. a teleprompter holds the most personal text on your screen. the unreleased pitch, the apology you're rehearsing, the thing nobody has seen. that has no business on someone else's server, and neither does the audio.
3. what happens when you go off script? ad-lib a sentence that isn't in the text, then land back on a written line. a good tracker holds position through the ad-lib and picks you up when you rejoin. a brittle one races off looking for a match.
4. what happens when you stop talking? silence should pause the scroll, not restart it and not run on. the pause is the thing that lets you scroll back, re-read a flubbed line, and continue. an app that keeps scrolling through your thinking time makes every mistake into a reshoot.
5. which microphone is it listening on? this one gets ignored and it decides whether the feature is usable on a real set. if the app only ever listens to the phone's built-in mic, voice following works at arm's length and nowhere else. step back to your mark and it goes deaf.
how open prompter does it#
voice tracking is opt-in, off by default, and labelled BETA. that label is honest, not modest. it works, and the feel is still being tuned.
the recognition is on-device. no audio leaves the phone, there's no account, and the app makes no network calls of its own, which you can confirm yourself because the source is public. the matching is loose on purpose, so a dropped article or an improvised aside doesn't throw it. silence for about 1.5 seconds pauses the scroll, and the page is yours during the quiet: drag back, re-read, and it meets you at the line you stopped on when you start talking again.
the feel is set by two draggable lines. the READ line is where the word you just said lands on screen. the FEATHER line is where the scroll hands off from snappy catch-up into a smooth glide. close together gives you tight, word-for-word tracking. spread apart gives you looser momentum. that one distance is the difference between a metronome and a current, and the READ and FEATHER guide walks through tuning it.
you write the script on your mac.
save the file. the phone catches up.
hit play, or let voice tracking follow you.
the READ line is where the word you just said lands.
the FEATHER line is where the scroll eases into a glide.
drag them apart for smoother momentum.
drag them together for snappy, word-for-word tracking.
no thumb on the screen. no subscription. no cloud.
you write the script on your mac.
save the file. the phone catches up.
hit play, or let voice tracking follow you.
the READ line is where the word you just said lands.
the FEATHER line is where the scroll eases into a glide.
drag them apart for smoother momentum.
drag them together for snappy, word-for-word tracking.
no thumb on the screen. no subscription. no cloud.
READ at 18% · FEATHER at 42% · band 24% (snappy follow)
on question five, it listens on the recording audio path. so it follows whatever microphone you feed the phone, not just the built-in one. route a lavalier in over USB-C or a 3.5 mm TRRS input, clip it to your collar, and walk back to your mark. the script keeps pace with your actual words from well past where the built-in mic would hear you. almost no other professional prompter app can track voice from an external mic source, and on a beam-splitter rig it's the difference between a prompter that works at arm's length and one that works at performance distance. the beam-splitter guide covers the wiring.
what it can't do#
worth saying plainly, because the failure modes are predictable.
it can't follow a script it doesn't have. voice tracking matches your speech against the text on screen. if you're improvising an entire section, there's nothing to match against and the page will hold rather than guess.
it can't hear you over a room. a loud fan, a busy street, three people talking at once. recognition quality is recognition quality, and no amount of matching logic rescues audio that never contained your words. this is the real argument for the wired lav, ahead of the range argument.
it doesn't write, edit, or summarise anything. the recognition exists to move the page and then it forgets what it heard. there is no cloud feature underneath it and no AI layer waiting to be turned on. open prompter doesn't generate scripts in-app, deliberately.
and it won't fix a script that's hard to say. words follow better when they're written to be spoken. that's a writing problem, not a prompter problem.
try it on a real take#
turn voice tracking on, read something you've actually written rather than a test paragraph, and deliberately fluff a line to see what the pause feels like. then drag READ and FEATHER until the scroll moves the way you read.
it's free, and there's no paid tier holding a better version of it. this is the version. it's also still in BETA, so if the tracking loses you somewhere specific, that's the useful thing to tell me.