Onwords support & FAQs.

Onwords is a teleprompter for Windows that follows your voice. You paste in a script, and as you read it aloud the app tracks where you are and scrolls the text to keep your current line centred on screen.

1. Paste your script into the box and click Start Prompting.
2. Choose your microphone from the Mic dropdown.
3. Click Start Listening and begin reading.

That’s Voice-Follow, the default. If you would rather have a steady scroll, switch to Auto-Scroll in the toolbar and the microphone controls give way to a Play button and a Speed slider in words per minute.

The line you’re reading is highlighted and sits on the guide line across the middle of the screen, with the lines above and below faded back. A marker sits under the word the app has reached.

Windows 11, or Windows 10 version 1809 or later, on a 64-bit machine — Intel, AMD or ARM — plus a microphone. Nothing else: no account, no subscription, no internet connection, and no runtimes to chase. The speech engines ship inside the app, which is why it is around 400 MB installed.

One feature needs more: the script rewrite uses the AI hardware in a Copilot+ PC, and stays hidden on machines without it.

Coming soon. Onwords is Windows first — a native Windows app with the speech recogniser bundled inside it, which is what lets it run entirely on your own machine, with no account and no connection. A Mac version is on the way. Join the mailing list and we’ll tell you when it lands.

£19.99, once (about $23.99), with 20% off at launch — so £15.99 while the sale runs. Most of the apps we tried wanted more than £10 a month, forever. Onwords costs less than two months of that, and then it is yours.

There is a free 7-day trial on the Microsoft Store with everything unlocked, so you can try it with your own script, your own microphone and your own accent first. There is no subscription and no account to create. The Store shows the price in your local currency.

No. A Microsoft Store purchase is tied to your Microsoft account, not to the machine. Sign in to the Store on the new PC and install Onwords from your library; it arrives already licensed. Microsoft’s current rules allow an app on up to ten of your Windows devices at once.

Your script does not travel with the licence — it is saved only on the computer you wrote it on — so copy that across yourself if you need it.

The first time you run Onwords, a short Getting started pane explains how it works. It does not appear again after that, but Help › Getting Started opens it whenever you want it.

Release notes work the same way: after an update the megaphone in the top-right corner opens by itself to show what changed. Click it any time, or use Help › What’s New.

Yes. Your script is saved on your computer automatically and restored the next time you open Onwords. You don’t need to paste it again between takes or sessions.

Long enough that it isn’t worth worrying about. The editor shows word count, character count and an estimated reading time as you type, and flags anything beyond about 5,000 words — roughly forty minutes of continuous speech — as unusually long for one sitting. That’s guidance, not a limit.

Not yet — scripts are pasted in, and whatever is in the box is saved for next time. Opening scripts from a file is on the roadmap.

The app listens through your microphone, converts your speech to text with a recogniser built into Onwords itself, and matches what it hears against your script. It looks at the last several words you have said and finds the place that best explains them.

That recogniser is a modern neural model running on your own processor. It was trained on real-world speech — podcasts and video as well as audiobooks — which is what lets it cope with accents, and with a presenter standing well back from the microphone. The classic Windows recogniser is still available under View › Recognition Engine if you prefer it.

Nothing breaks. The app needs only a few recognisable words to keep its place, so a fluffed line, a filler word or a short unscripted aside won’t throw it off. When it isn’t confident, it holds position rather than guessing.

It will catch up. When what you’re saying stops matching the area it’s watching, the app searches your whole script and jumps to where you’ve moved to. This usually takes a sentence or two.

No — that’s deliberate. The app waits until it’s reasonably confident before it moves, rather than leaping ahead on a single word it might have misheard. In practice the marker sits a word or two behind your voice, which is what keeps it steady enough to read from. The highlighted line is what you follow; the marker is a finer cue.

Usually not. That bar shows the recogniser’s raw output, and even a good recogniser mangles the odd name or technical term. Tracking doesn’t need a perfect transcript — only enough recognisable words to locate you in the script. If the transcript looks rough but the highlight is following you correctly, everything is working as intended.

The pill at the right of the bar names the engine that is listening.

• Use a decent microphone positioned close to you, rather than a laptop or webcam mic. A wireless lavalier is the single biggest improvement for a presenter who moves about
• Record somewhere reasonably quiet
• Speak at a natural, even pace rather than rushing
• Try the other engines under View › Recognition Engine — see below

Onwords already lifts a quiet microphone automatically, so your distance from the laptop matters far less than it used to.

Yes, under View › Recognition Engine. The default is a neural recogniser tuned for general speech. A second neural engine is biased toward the words in your own script, which can help with unusual names. The classic Windows recogniser is there as well, and if you use that one, training a Windows speech profile (Settings › Time & Language › Speech) tunes it to your voice.

The menu is locked while the app is listening, because changing engine mid-take would restart recognition under you.

English. The speech engines bundled with Onwords are English, and English is what we support.

If you present in another language you can try View › Recognition Engine › Windows (classic), which uses whichever recogniser Windows has installed for your display language — French and German ship with Windows, and others can be added through Settings › Time & Language › Language & region. We have not tested those and cannot promise how well they track.

Yes. Tick Two speakers under the script box, select the lines belonging to one presenter, and click Speaker A or Speaker B. In the prompter, each speaker’s lines are tinted — pale yellow for A, pale blue for B — so you can see at a glance whose line is coming next.

Assignments are stored in the script itself as a short prefix (A: / B:), so they survive editing, they’re saved with your script, and you can type them yourself instead of using the buttons if you prefer. The labels are shown in the prompter but are never treated as words to be spoken — Onwords won’t sit waiting to hear you say “A”.

No — and it won’t use two. Both presenters share one microphone. Onwords follows whoever is speaking, so it tracks straight through the handovers. If the second person is off-mic or across the room, it won’t hear their lines and the highlight will sit still until someone it can hear resumes.

Yes. Click Mirror to flip the text horizontally for the glass.

Use the A- and A+ buttons to suit your distance from the screen.

Yes — drag the window to your second display and click Fullscreen, or use View › Toggle Full Screen. The controls stay where you left them.

Yes. In the prompter, switch the mode from Voice-Follow to Auto-Scroll. The microphone controls are replaced by a Play/Pause button and a Speed slider, so the script sits still until you press Play and you set the pace in words per minute. Restart takes you back to the top and pauses, ready for a cue.

On a Copilot+ PC, yes. A small wand button appears beside the script box. Pick a style — Neutral, Casual, Concise, Formal, or your own instruction such as “rewrite as a warm, upbeat voiceover” — and Onwords shows you the result before changing anything. Nothing is replaced until you say so.

This runs entirely on your device, using the language model built into Windows. Your script is not uploaded anywhere.

The button is hidden on PCs without the hardware for it, and in two-speaker mode, where a rewrite would strip the A: and B: labels your script depends on. Note that it shortens as it works — expect roughly 15% fewer words — so read the preview before you replace anything.

Click ↻ Restart. This returns to the beginning of your script and clears the live transcript.

Yes. Use the ↑ and ↓ arrow keys to move the highlight a line at a time. You can also scroll with the mouse wheel at any point.

No. Your script stays on your own machine. It’s stored as an ordinary text file in your own user folder, and you can delete it at any time.

Entirely on your computer. The recogniser is bundled inside the app and runs on your processor; nothing is uploaded, no online service is involved, and no audio leaves your machine. The optional script rewrite runs on the device too.

No. Audio is processed as it arrives and is never written to disk, and Onwords only listens while you have pressed Start Listening.

There is one exception, and you have to switch it on yourself: Help › Recognition Logging writes the recognised text — not audio — to a local file, to help us with support. It is off unless you turn it on.

Yes, completely. Onwords needs no connection at any point, and makes no network calls of any kind.

No. There’s no sign-in, and no account to create.

This is nearly always the microphone selection. Windows’ default recording device is often not the microphone you’re actually speaking into — it might be a webcam, a virtual audio device, or a built-in array pointed at the wrong thing, and the signal from it can be far too quiet for speech recognition to work with.

Open the Mic dropdown and pick your actual microphone by name, then click Start Listening again. If the “Heard” bar starts showing words, it’s working.

Onwords will usually tell you when this is the problem. While it’s listening, a small level meter sits next to the Mic dropdown — it should move as you speak. If nothing usable arrives for a few seconds, an orange No audio detected bar appears across the top of the script. That message means the microphone, not the tracking.

If the No audio detected bar is showing, work through these in order:

1. Check the Mic dropdown is set to the microphone you’re actually speaking into, not “System default”
2. Check Windows can hear that microphone: Settings › System › Sound › Input, and watch the input level as you speak
3. Confirm Windows lets desktop apps use your microphone: Settings › Privacy & security › Microphone
4. Click Stop Listening, then Start Listening again

Scripts that repeat a phrase in several places can occasionally cause this. Use the arrow keys to move the highlight back to the right line, or click ↻ Restart and carry on from the top. Continuing to read will normally also bring it back on its own.

Stop listening first, choose the new device, then start listening again. The dropdown is locked while the app is listening, and so is the recognition engine menu — changing either mid-take would restart recognition under you.

Turn on Help › Recognition Logging. From then on, each listening session writes what the engine heard, and where the marker moved, to a plain text file at %LOCALAPPDATA%\Onwords\recognition-log.txt.

Help › Open Log Folder takes you straight there and shows how much is on disk. It grows by roughly 30 KB per minute of speech and is capped at 5 MB, with the oldest sessions rolling into a .previous file. It is off by default because it contains what was said — the one thing Onwords otherwise never keeps — so switch it off again when you are done.

The full FAQ documentation goes into more detail than this page. If it isn’t answered there, get in touch and tell us what you were doing when it happened — and if it is about voice tracking, a recognition log (above) will help us see what went wrong.