Voice-Controlled Teleprompters: How Speech-Following Scroll Works

A traditional teleprompter scrolls at a fixed speed. You adapt to it. A voice-controlled teleprompter flips that: it listens and follows you. For anyone who pauses, gestures, or improvises, this is the single biggest upgrade in how prompters work.

How it works, in plain English

1. The microphone streams audio to a speech recognition model.
2. The model produces words as you say them.
3. A matcher compares those words with your script to find where you are.
4. The display moves so your current line stays in the reading spot.
5. If confidence drops (you ad-lib, cough, or someone talks), it holds position instead of guessing.

Step 3 is what makes this reliable. The app is not trying to understand anything you might say; it only has to find your place in text it already knows. That is a much easier problem than open-ended transcription.

Why on-device AI changed things

Early voice prompters sent audio to the cloud, which added lag and privacy concerns. Compact speech models now run on a phone's own processor in real time. Cue's Voice Follow uses an English streaming model bundled in the app: no audio leaves the device, and audio is discarded as it is processed.

The benefits:

• Low latency: no network round-trip
• Works offline: basements, planes, studios
• Private: your voice is never uploaded

When voice-follow beats fixed speed

• You pause for effect or laughter
• You gesture or demonstrate something mid-sentence
• You paraphrase rather than read word for word
• You are on a rig and cannot reach the screen
• You are nervous and your pace changes through the take

Fixed speed still works well for steady, practiced reads, and some people prefer its predictability.

Tips for reliable tracking

• Quiet room. Background talk and music confuse any recognizer.
• Mic close to you. A lapel mic or the phone within arm's length helps.
• Write distinctive lines. Repeated phrases ("and so", "and so") can make the matcher jump. Vary them.
• Start on the first line. Give the recognizer a clean anchor.
• Tune the hold setting. Cue lets you adjust how much confidence it needs before moving, so you can trade responsiveness for stability.
• Keep directions out of speech. Stage directions like [pause] are not counted as words in Cue, so the matcher only looks for what you will actually say.

Limitations to know

• It follows the language it was trained for (Cue's current model is English).
• Skipping far ahead may need a manual jump, and sections help. In Cue, # headings create a jump list.
• Very noisy environments reduce accuracy.

Try it

If you have struggled to match a scroll speed, voice-follow is worth trying before you give up on prompters altogether. Pair it with the eye contact techniques and a script written for the ear and you will sound like you are simply talking.