Recording to camera in your second language.

Why it is harder than it should be

Speaking a second language well in conversation and reading it aloud to a lens are different skills. In conversation you choose your words as you go, so you only ever use the ones you own. A script was written at a desk, with time to reach for the better word, the longer phrase, the idiom you read somewhere. On camera you have to pronounce all of it, at speaking speed, with no one in front of you to nod.

Three things happen at once:

• Your reading speed drops. You decode each word a fraction slower than in your first language. Not much, but it adds up across a line.
• Your pace becomes uneven. Easy words go fast. A hard word, a name or a number makes you slow down or stop for a beat while you get your mouth around it.
• Your eyes work harder. You check the next word before you say it, more often than you would in your own language. That checking is visible.

None of this means your accent is the problem. An accent is fine, and often an asset: it tells the viewer who they are listening to. What reads badly is the stop-start rhythm and the eyes that never settle. Those can be fixed.

Write the script you can say, not the one you can write

Most of the work happens before the camera is on. The goal is a script made only of words you can say without thinking about them.

• Use the words you use when you talk. If you would never say "furthermore" in a meeting, do not write it. Your spoken vocabulary is smaller than your written one, and that is the one the camera hears.
• One idea per sentence. Long sentences with a clause in the middle are hard in any language. In a second one, you lose the thread halfway and the end comes out flat. Cut them in two.
• Read it out loud once, slowly, and circle every stumble. Each circle is a word to replace or a sentence to rebuild. Do not try to "get better at" a word the night before; swap it for one you already own.
• Watch out for words that look alike. Words that are spelled almost the same in your two languages are where the wrong pronunciation sneaks in. If one keeps catching you, pick another word.
• Break lines by meaning. Short lines, one phrase each. You find your place faster when you look back, which matters more when you look back more often. How to format a script for a teleprompter covers the layout in detail.

A good test: if you could say the line to a colleague without looking at it, it belongs in the script. If you need to read it twice to say it once, it does not.

Pace is the real problem

Here is the part that is specific to a second language. In your first language, your speed already varies a lot inside one script: slower on numbers, faster on what you know well. That is the reason a fixed-speed scroll makes anyone sound like they are reading.

In a second language, that variation is much bigger. A sentence you have said a hundred times goes by quickly. Then a word you rarely use arrives, and you need an extra second to land it cleanly. A timer prompter does not know that second exists. It keeps going, the next line slides up while you are still on the hard word, and now you have two choices: rush the word and get it wrong, or stop and lose the line.

Setting the scroll slower is the usual answer. It helps with the hard words and ruins the easy ones, because now you are waiting on sentences you could say twice as fast. You end up filling the gaps, stretching vowels, or getting ahead of the text and stalling mid-sentence. The take sounds careful and slow, which is the opposite of what you wanted to show.

The honest conclusion is that no speed is right, and the more uneven your reading is, the more wrong any fixed number becomes.

A scroll that waits for you

The fix is to stop asking the scroll to guess. You can do that with any prompter:

• Record one short chapter per take. A few lines at a time, then cut between them in the edit. A hard word only costs you one short take, not the whole script.
• Keep a pause key or a remote in reach. Hold the scroll on the hard word instead of slowing the whole script down for it.
• Set the speed for your easy sentences. Let the scroll run at the pace you say the words you own, and pause by hand on the few you do not.

The next step is a scroll that does this for you. If the text moves because you said the words, a hard word costs you nothing: you take the extra second, the text waits, and it moves on when you do. The easy sentences go as fast as you say them.

This is what Voice follow does in Mirrento. It listens through the Mac's own speech recognition and moves the text only when what you say matches the script. While Voice follow is on, the word you are on is highlighted, what you have read is dimmed, and the view moves in blocks rather than a constant crawl, so the lines in front of you stay still while you read them. For someone checking the next word more often than usual, that matters: when your eyes come back, the line is where you left it.

It also changes what a stumble costs. If you trip on a word and say the phrase again, it follows you back. If you stop to think, nothing moves. You are not racing anything, which is most of what makes second-language takes sound tense.

Two limits, said plainly. Voice follow matches what you say against what is written, so a word you pronounce very differently from the way the recogniser expects can slow it down for a moment. It catches up when the next words match. And it works best in a quiet room with one voice, as any speech recognition does.

Any voice-driven prompter has to listen in the language you are speaking. Set it per script, and install your system's on-device language model the day before, not the morning of. Voice commands work the same way: give each one a phrase in the language you record in, so you can say "next" without switching languages mid-take. Setting up voice commands in your own words covers how.

If you record in two languages and the scroll is what keeps tripping you, the 7-day trial is enough to record the same script with a fixed scroll and with Voice follow, and hear which one sounds like you.