Voice-following vs fixed-speed scrolling: how each works
Quick answer
Fixed-speed scrolling moves the script at a constant rate set by you, so it is predictable but cannot react to pauses. Voice-following scrolling uses speech recognition to find the sentence you are saying and moves the script to match, so it waits when you wait but depends on a quiet room and language support.
A teleprompter has to decide when to move the text. Fixed-speed scrolling moves it by the clock. Voice-following scrolling moves it by listening to you. Both are useful, and each fails in its own way. This guide explains the mechanics of each, the errors they produce and how to choose, with Telepromptia as the worked example.
Key takeaways
- Fixed speed is a clock: the position of the text depends only on time and on the speed you set.
- Voice following is a feedback loop: the position of the text depends on the words you say.
- Fixed speed cannot react to pauses or slips. Voice following can, but it needs a microphone, a quiet room and a language the recognizer supports.
- Fixed speed can make a script fit an exact length. Voice following cannot, because you set the pace.
- Many people use both: voice following first, and a fixed or manual scroll as the fallback.
How does fixed-speed scrolling work?
The app moves the text up the screen at a constant rate. You choose the rate, usually in steps, and it multiplies the time since you pressed start. Nothing listens to you. Telepromptia can also work backward from a duration: if you pick a target length of 30, 60 or 90 seconds, the app divides the length of the script by that time and sets the speed so the script lasts that long. If the script cannot fit, it tells you how many words to cut.
Manual scrolling is the same idea with your finger as the clock: full control, but it needs a free hand.
Where does fixed speed go wrong?
The weak point is that people do not speak at a constant rate. In a University of Pennsylvania analysis of recorded English telephone conversations, rates ran from about 111 to 291 words per minute, and even one person speeds up on easy lines and slows down on hard ones.
The error also builds up. Suppose the scroll runs at 140 words per minute and you read 10 percent slower. After one minute you are about 14 words behind, and after three minutes about 42. A pause makes it worse, because the text keeps moving while you think. The usual defense is a steady rhythm, which tends to sound like reading.
How does voice-following scrolling work?
It is a loop with four steps.
- Capture. The app listens through the microphone.
- Recognize. A speech recognizer turns the audio into words. Recognizers can report partial results while you are still speaking, so the app does not wait for a full sentence.
- Align. The app compares the recent words with the script near the current position, ignoring capitalization and filler words, and decides which sentence you are on. A matcher that only looks near the current position is fast and avoids jumping to a similar phrase far away.
- Move. The script scrolls so the current sentence sits in the reading zone. If nothing matches, such as a cough or a side remark, the position holds.
There is a short delay between your words and the result, so the script follows you rather than leading. Good implementations smooth the movement so the text glides instead of jumping.
Where does voice following go wrong?
- Noise and other voices. Background speech can match words in the script by chance, or hide yours.
- Unusual words. Names and numbers can be recognized differently from how the script spells them. Write them as you say them.
- Language support. A recognizer works only for the languages and devices it supports.
- Large skips. If you jump far ahead, the matcher has to be confident before it follows.
This is why a good app holds its position when it is unsure, instead of guessing.
Where does the speech recognition run?
On Apple devices, an app can require recognition on the device and check whether the device supports it for a language. Telepromptia's FlowSync uses on-device recognition only and never falls back to a server, so your voice is not sent anywhere for this feature and it works without a connection. If a language has no on-device support, FlowSync is not offered for it.
How do the two compare?
| Criterion | Fixed speed | Voice following |
|---|---|---|
| What moves the text | A clock and your speed setting | The words you say |
| Pauses | The text keeps moving | The text waits |
| Skipped or repeated lines | No recovery | Recovers once it is confident |
| Exact video length | Yes, with a target length | No, you set the pace |
| Noisy room | Not affected | Can be affected |
| Language support | Any language | Where on-device recognition exists |
| Processing | Almost none | Speech recognition runs while you read |
Which should you choose?
- Choose fixed speed for an exact length, a noisy room, a language without on-device recognition or a script you have rehearsed until your pace is even.
- Choose voice following for longer or less rehearsed scripts, or when you like to pause and think.
- Choose manual scrolling for a short script, or when you want to control every line.
Try each on the same 150-word script and see which one you stop noticing. Our guide on font size and speed shows how to measure your own pace.
Where Telepromptia fits
Telepromptia has both. Fixed-speed scrolling, a target length, a countdown, a loop and manual scrolling are in the free plan. Voice-following scroll (FlowSync) is part of Pro. Scripts and videos stay on your device, with no account. See how voice following works, the features page and the pricing page.
Common questions
Is voice-following scroll the same as voice commands?
No. Voice commands react to words such as start or stop. Voice following listens to the words of your script and keeps the text in step with you.
What happens when I pause or say something that is not in the script?
The script waits when you pause. When the app hears something it cannot match, it holds its position instead of guessing.
Does voice following need an internet connection?
Not in Telepromptia. It uses on-device speech recognition only, so it works offline, and it is simply not offered for a language the device cannot recognize on the device.
Which is better for a one-minute video?
Fixed speed with a target length is often enough, because a short script is easy to rehearse. Voice following helps more as scripts get longer.
Sources
- supportsOnDeviceRecognition (Apple Developer Documentation): whether the speech recognizer can operate without network access.
- requiresOnDeviceRecognition (Apple Developer Documentation): a request can be set to keep its audio data on the device.
- shouldReportPartialResults (Apple Developer Documentation): a recognition request can report results before the speech is finished.
- Yuan, Liberman and Cieri (2006), Towards an Integrated Understanding of Speaking Rate in Conversation (University of Pennsylvania): in recorded English telephone conversations, speaking rates ranged from about 111 to 291 words per minute.