About this tool
Automatic transcripts from meetings, lectures, podcasts and voice notes usually arrive as one long, lower-case stream of words. Reading it is slow and quoting it is painful. This tool sends each passage through an instruction-tuned multilingual model that inserts periods, commas, question marks, exclamation marks and apostrophes, and capitalizes sentence starts — following each language’s own conventions, such as Spanish ¿…? or French spacing. Your words are never rewritten: a validation step compares the letters before and after and falls back to rule-based punctuation if anything else changed.
Like every tool on LanguageHelper, AI Punctuation Restorer runs entirely in your web browser. Your text is never uploaded, stored or used for training, which makes it safe for confidential documents, client data and personal messages. No account or installation is needed, and the tool keeps working offline once loaded.
The first run downloads a compact AI model — about 320 MB for the default fast model, or 480 MB for the best-quality one. It is stored in your browser, so later runs start in seconds and work offline. An instant rules-based result is shown while the model loads; switch the engine to “Rules” to skip the download entirely.
How it works
- Pick the language of the transcript, or keep auto-detect.
- Paste the unpunctuated text — line breaks are preserved.
- Click “Add punctuation” and wait for the model to process each chunk.
- Copy or download the result, or compare it with Text Diff.
Use cases
- Make meeting, interview and lecture transcripts readable.
- Prepare speech-to-text output for subtitles and captions.
- Punctuate voice notes and dictated messages before sending.
- Clean up chat logs or OCR text before analysis or translation.
Examples
Before
dr smith moved to new york in 2019 she loves cafe culture doesnt she
After
Dr. Smith moved to New York in 2019. She loves cafe culture, doesn’t she?
Before
sabias que mañana ira a bogota
After
¿Sabías que mañana irá a Bogotá?
Tips for best results
Keep one speaker turn per line when you can: line breaks are preserved and give the model natural boundaries. Remove timestamps and speaker labels before running if they are not needed, then paste them back later. After punctuating, run the Sentence Splitter to get one sentence per line for subtitle editors or translation tools, and use Text Diff to confirm that only punctuation changed.
Frequently asked questions
Does it work for languages without spaces?
Yes. Chinese and Japanese get full-width punctuation (。、?), Thai gets sentence spacing, Hindi uses the danda (।) and Arabic uses its own question mark and comma (؟ ،). The model handles all 20 site languages.
How long can the text be?
There is no hard limit. Long texts are split into chunks of a few sentences and processed one after another, with the output streaming into the result box as it is generated.
Will it fix spelling as well?
No — by design it only adds punctuation and capitalization so your transcript stays faithful to what was said. Use the Spell Checker or Common Mistake Fixer afterwards for spelling.
Is it really private?
Yes. The AI model runs inside your browser tab using WebGPU or single-threaded WebAssembly. No audio or text is sent anywhere, which makes it suitable for confidential meeting notes.
What is the rules engine?
Rules mode adds a final period or question mark to each line based on question words, inserts commas before common conjunctions and capitalizes sentence starts. It is instant and a good fallback on slow devices.