Updated September 2026
The homepage's "How it works" section covers the idea in a sentence: Voice Follow listens as you read and scrolls to match your pace. This page goes one level deeper — the actual mechanics behind that, in plain terms, so you understand what it's doing (and why it sometimes seems to "guess") instead of treating it as a black box.
The moment you click Start — Voice Follow, your pasted script is split into individual words. That word list is the thing the tool is actually tracking a position in — not the raw text, not sentences, not paragraphs. A "read pointer" starts at word zero and only ever moves forward.
This tool doesn't run its own speech-recognition model. It uses the Web Speech API built into your browser (SpeechRecognition / webkitSpeechRecognition) in continuous, interim-results mode, so it gets a steady stream of "here's what I think I just heard" results rather than waiting for you to finish a whole sentence. In Chrome and Edge, that recognition typically happens by streaming your mic audio to Google's speech servers and getting text back — see the privacy policy for the full, honest breakdown of what that means for your audio.
This is the part that makes Voice Follow usable in practice. Instead of demanding that what the browser heard exactly matches the very next script word, the tool compares each recognized word against a window of the next several upcoming script words — not just word #1, but a short lookahead range past your current position. Within that window, a match doesn't have to be a perfect character-for-character hit either: for words of four or more letters, it accepts a fuzzy match (an edit distance of one — a single letter swapped, dropped, or added), which absorbs the kind of small transcription slips speech recognition makes constantly.
When a confident match is found somewhere in that window, the read pointer jumps to just past that word — not to the next word in sequence, but to wherever in the window the match actually landed. That's deliberate: if you skip a line, ad-lib a phrase, or the recognizer simply drops a word it heard, the pointer can still catch back up to where you really are instead of stalling on a word that never gets confirmed.
Not every word the recognizer reports will land a match — filler words, false detections during background noise, or a phrase that doesn't appear verbatim in your script all show up as "no match in the window," and the tool just discards them and keeps listening. Nothing gets logged as an error, nothing pauses the session; the pointer only moves when something in the window is confirmed.
Once the pointer advances, the script view smooth-scrolls so your current word sits roughly a third of the way down the screen rather than pinned at the top. That leaves a few lines of "what's coming" visible above your eye-line and puts your current line at a natural reading position instead of forcing your eyes to the very edge of the display.
Chrome and Edge's continuous recognition mode doesn't actually run forever — it ends the session after a gap of silence, which is a browser-level behavior this tool doesn't control. When that happens mid-read, the tool detects the session ending on its own and immediately restarts a new one behind the scenes, as long as you haven't explicitly hit pause or stop. In normal use this is invisible; you just keep reading and it keeps listening.
Because matching is fuzzy and window-based rather than exact, you don't need to read your script word-perfect for Voice Follow to keep up — small ad-libs, a skipped filler word, or a stumble you correct out loud are all tolerated. What trips it up more than misreads is going completely silent for an extended stretch (which can end a recognition session before the auto-restart kicks back in) or jumping to a part of the script far outside the lookahead window, which is genuinely outside what window-based matching can find. If you expect to jump around a script non-linearly, Manual Scroll Speed mode — which just auto-scrolls at a speed you set, with no listening at all — is the more predictable choice.