How a voice-activated teleprompter works

How the tracking works

The app already knows every word of your script, which turns a hard problem into an easy one. It does not need to understand open-ended speech. It only needs to work out where in a known text you are, and it does that by matching the words it hears against the words it expects next.

That is why tracking stays accurate through unusual names and technical terms that defeat general dictation. The recogniser does not have to spell the word correctly. It only has to notice that you said something in the right place.

What it copes with, and what it does not

• Pausing is handled completely. The script stops when you stop, for as long as you like.
• Going off script for a sentence is fine. The scroll holds its position and picks up again when you return to the text.
• Skipping ahead works if you land on a distinctive phrase, and less well if you land on common words.
• A noisy room is the real enemy. A fan, music or a second voice degrades tracking faster than anything you do.

If tracking drifts, the cause is almost always audio rather than the app. Move the microphone closer before you change any setting.

Getting a clean first take

1. Check that the recognition language matches the language of the script. A mismatch produces tracking that almost works, which is harder to diagnose than tracking that fails.
2. Use an external microphone if you have one. A Bluetooth microphone near your mouth outperforms a built-in one across the room.
3. Read the first line at your normal volume. Tracking locks on within a few words, and starting quietly delays that.
4. Leave the manual controls within reach. You can take over the scroll at any moment, and hand it back.

When it is the right tool

Voice tracking suits anything conversational: interviews, tutorials, pieces where you want to react and elaborate. It suits you less if the piece must hit an exact duration, because the script no longer sets the pace. For that case see voice-activated or fixed speed.