Teleprompter languages: 60 languages, one prompter that follows your voice

How language support works

Three things happen on every take, and they are the same in every language. Speech recognition turns your voice into words, marking each word with the language it detected. The matcher compares those words with your script after normalising both sides: accents and other diacritics are removed, everything is lowercased, and punctuation is dropped. Then the scroll follows the matched position — waiting when you pause, catching up when you re-read a line, and re-locking when you return from an ad-lib. The rules that decide whether a spoken word matches a script word are the same everywhere:

• Short words must match exactly. Words under five letters need an exact match after normalisation, which is why articles and particles are carried by their neighbours rather than matched on their own.
• Longer words tolerate small differences. Five letters or more allow one letter of difference; eight or more allow two. A dropped ending or a recogniser slip inside a long word does not stall the prompter.
• Diacritics never matter for matching. "año" and "ano", "schön" and "schon", "qué" and "que" all match the same spoken word. Keep the marks for your own reading; they cost nothing.
• Hyphens and apostrophes split words into pieces that are matched one after another: "anak-anak", "Türkiye'de", "l'ordinateur" each consume their parts in order.
• Numbers are matched as written. The recogniser often writes digits while your script has words, or the other way round, and there is no conversion — so write every number the way you will say it, in any language.

Matching runs in your browser on the live transcript, and the script itself never leaves your device. For the mechanics of pauses, re-reads and skips, see how a voice-controlled teleprompter works.

All 60 languages, grouped by script

Grouping by script is more useful than an alphabetical list, because the script — not the grammar — decides how well word-by-word following works today. Every language below is transcribed; the tables note what to expect from the matcher and link to the dedicated guide or localised site where one exists.

Latin script, spaced words: full word-by-word following

Cyrillic script: full word-by-word following

Greek: full word-by-word following

Right-to-left scripts: full following, rendered right to left

Arabic, Persian, Urdu and Hebrew are transcribed and followed word by word. In the prompter every paragraph takes its own direction, so a right-to-left paragraph reads from the right edge with punctuation where it belongs while an English paragraph in the same script stays left to right; the editor follows the direction of the script as a whole. Mirroring for beam-splitter glass flips the entire scroll area, direction included.

Indic scripts: following works, with one caveat

Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Punjabi, Tamil and Telugu are transcribed and followed. The caveat: when the matcher normalises a word it keeps letters and digits only, and in these scripts the vowel signs that sit around a consonant are marks, not letters. Matching therefore compares the consonant skeletons of words. Two short words that differ only in their vowel signs look identical to the matcher, and the surrounding words are what keep the position right. In practice tracking is stable through normal sentences; where it is most likely to hesitate is a line of very short, similar words. Write each language in its own script — the transcript comes back in the script the recogniser uses for that language, Devanagari for Hindi — and keep English words inside the same line in Latin letters, where they match as English.

Hangul: full word-by-word following

Chinese, Japanese and Thai: transcribed, not yet followed word by word

These three languages are recognised and transcribed, but voice following is not available yet. The matcher works on words separated by spaces, and these scripts write sentences without word boundaries, so the prompter cannot line up what you say with the words on screen. Adding spaces to the script does not help on its own, because the transcript comes back without word boundaries. Word segmentation for these scripts is on the roadmap; until it ships, the honest answer is that Sayscroll is not the right prompter for a Chinese, Japanese or Thai script.

Mixed-language scripts

Most real scripts are bilingual somewhere: an English product name inside a Hindi review, a quoted headline, a Spanglish intro for a US Latino audience, Taglish, Denglish. Sayscroll transcribes each word in the language you say it and matches it to the script, so the switch never breaks tracking and there is no toggle mid-take. Two rules make it reliable:

• Write each word the way you will say it, in its own script. A brand name pronounced in English is written in Latin letters even in the middle of a Devanagari or Arabic line.
• Keep the hint in mind. On the English site the recogniser is hinted toward English with the other language identified from context; on a localised site the hint is reversed. Both directions follow both languages — the hint only decides which language wins a close call on a word that exists in both.

The six dedicated guides go deeper: Spanish, Portuguese, French, German, Italian and Swedish, plus the English teleprompter page for accents on the English side.

Writing a script the prompter reads well, in any language

Recognition transcribes what it hears, so spelling that mirrors speech tracks better than formal written language. The same five habits work in all 60 languages.

• Numbers as words, including years, prices and percentages, because digits and number words are never converted.
• Abbreviations as spoken: write "for example" or its equivalent rather than "e.g."; write an acronym as a word if you say it as a word, and as hyphenated letters if you spell it out.
• Loanwords and names in the language you will pronounce them, not the language of the surrounding sentence.
• One thought per line. Short lines keep the active word near the lens and make a re-entry after an ad-lib instant, because the matcher only has to find you within a few words.
• Pauses as punctuation, not stage directions. A full stop or a dash is enough; "(pause)" is just another word to the matcher and does not create a pause.

Languages with their own site, and guides in English

Six languages have the whole product, blog and free tools localised, with the recogniser hinted toward that language first: Español, Português, Français, Deutsch, Italiano and Svenska. For English-speaking creators who record in one of those languages, the dedicated guides above cover varieties, mixed scripts and writing tips. Every other language on this page runs on the English site with automatic language identification — the normal case for bilingual creators, language teachers, and teams producing a video in a language they read but do not run their tools in.

Try it in your language

Paste a short script in any of the languages above on the free teleprompter page and press Present — no account, download or card. The language is identified from your first words; if you record mostly in Spanish, Portuguese, French, German, Italian or Swedish, start from that language's site so the hint points the right way. Everything else — pauses, re-reads, ad-libs — behaves as described on the voice-controlled teleprompter page, and you can check your pace first with the WPM reading test.