How a teleprompter that follows your voice works

Checked on 2026-10-26 against EyePrompt for iPhone and iPad 1.6.2, EyePrompt for Mac 1.1.1 and Apple's Speech framework documentation. We make EyePrompt, so weigh what follows with that in mind.

A teleprompter can move the script in three ways: at a fixed speed, when it hears you speak, or by following the words you say. Following the words is the most useful for a take or a talk, and the hardest to get right. This page explains how each way works, what it costs and where it fails.

The three ways a teleprompter can move

WayWhat it listens toWhen you pauseWhere it falls short
Fixed speedNothingKeeps scrolling until you stop itYou read at the pace of the machine, so you chase the text or the text waits on you.
Loudness (EyePrompt's Smart Pause, on iPhone and iPad)The level of the microphone, nothing elseHolds when you stop talking and starts again when you speakIt does not know where you are in the script. If you ad-lib or skip a line, the text still moves at the speed you set.
Words (EyePrompt's Smart Follow)The words you say, recognized and matched to your scriptWaits, then catches up when you skip aheadNeeds speech recognition for the script's language. A word is reported a moment after you say it, and noise can make it lose the thread.

What Smart Follow does

Moves the script at the pace you speak, waits while you pause or ad-lib, and catches up when you skip ahead. Speech recognition runs on the device, in the language the script is written in, and is never sent to a server. On by default on Mac, and on iPhone and iPad from version 1.4, when the device's language supports on-device recognition. From version 1.6.1 on iPhone and iPad, it starts with your voice, not a breath or a noise, and no longer jumps ahead after a pause. On Mac, from version 1.0.6, a recognized word moves the band only when it continues the script from your place, so other people talking on a call do not move it. From version 1.6.2 on iPhone and iPad, it finds your place again after mishearing a few words and keeps going, the recording screen shows which language it is listening in (guessed from the script), and you can pick the language and accent there.

It listens through the microphone the app is already using, turns what it hears into words on the device, and looks for those words in your script. It uses recognition to find your place, not to write down what you said.

On the device or in the cloud?

Speech recognition can run on a server or on the device. Apple's Speech framework lets an app ask which one is possible: a recognizer reports whether it can work without network access, and a request can be required to keep its audio on the device. Apple notes that on-device recognition is less accurate than the server's.

EyePrompt asks for on-device recognition only. If the device has no on-device model for the language of your script, Smart Follow does not start, and nothing is sent to a server instead. The cost is accuracy. EyePrompt offsets part of it by working from a script it already has: it only has to find your place in known words, so it does not need a perfect transcript. The benefit is that your voice stays on the device and the app works in Airplane Mode.

Why the script trails your voice

A recognizer can only report a word after it has heard it. Apple's API can also return partial results, intermediate guesses for each utterance that are revised as more audio arrives. A follower that chases the last recognized word therefore runs behind you, and stops when recognition stalls.

EyePrompt does not chase the last word. It glides at your estimated pace and treats each confident match as a correction, nudging the script toward a couple of words ahead of the word it matched. That small lead keeps the words you are about to say, rather than the words you just said, under your eyes.

When it fails, and what happens then

Getting a good result

Where it runs

Smart Follow is part of the free plan on iPhone, iPad and Mac. It is on by default on Mac, and on iPhone and iPad from version 1.4, where the device supports on-device recognition for the language. The app is translated into nine languages, and Smart Follow works in any language the device can recognize on the device. On a Mac, the same band reads your notes during a video call.

Download on the App Store

How EyePrompt works · How to look at the camera while reading a script

Sources: Apple Developer Documentation, SFSpeechRecognizer.supportsOnDeviceRecognition, SFSpeechRecognitionRequest.requiresOnDeviceRecognition and shouldReportPartialResults. The EyePrompt behaviour described here comes from the app's own settings screens and code, checked on 2026-10-26.

Questions people ask

Does a teleprompter that follows my voice send it to a server?

It depends on the app. EyePrompt's Smart Follow uses on-device recognition only, so no audio is sent anywhere and the app works in Airplane Mode. If a language has no on-device model on your device, Smart Follow is unavailable for it rather than falling back to a server.

What is the difference between Smart Follow and Smart Pause?

Smart Pause listens to loudness only: it holds the script when you stop talking and starts it again when you speak, on iPhone and iPad. Smart Follow recognizes your words and matches them to the script, so it also knows where you are. Both are in the free plan.

What happens if I skip a line or ad-lib?

Smart Follow waits while you ad-lib and catches up when it hears words from further down the script. If it cannot tell where you are, it holds instead of scrolling to a wrong place.

Does it follow my voice in my language?

In the language your script is written in, when your device can recognize that language on the device. The app is translated into nine languages, and Smart Follow is not limited to them.

Related guides

All guides

Google lets you pick the sources you want to see more of. You can add EyePrompt as a preferred source on Google.