Auto-captioning a video for free — how accurate is browser-based transcription, really
October 29, 2026 · 4 min read
Captions help watch-through rates, accessibility, and silent autoplay on social feeds — but paid auto-captioning services charge per minute, and manually typing them out for a five-minute clip is nobody's idea of a good afternoon.
What's actually happening when a video gets auto-captioned
Whisper is an open-source speech recognition model trained on a large, multilingual set of audio paired with its transcript. A browser-based captioning tool can run it directly on your device via WebAssembly — the model downloads once, then every video after that is transcribed locally, without audio ever being sent anywhere.
Where it does well, and where it doesn't
- Does well: a single clear speaker, common languages and accents, and a relatively quiet background — the conditions most talking-head and tutorial-style video actually has.
- Struggles: loud background music under dialogue, multiple people talking over each other, and uncommon proper nouns, brand names, or jargon the model hasn't seen much of.
- Accuracy on less-represented languages and strong regional accents is real but noticeably behind what it achieves on clear, standard-accent English or other high-resource languages.
Getting a cleaner result
- Trim out long silent or music-only stretches first — less audio for the model to guess at means fewer chances to introduce an error.
- If a clip has heavy background music, expect more mistakes around quieter dialogue and plan to spot-check those sections rather than trusting the output blindly.
- Always skim the result before publishing — auto-captions are a very strong starting draft, not a guaranteed-correct transcript.
Doing it without a subscription
Add Captions to Video auto-transcribes and burns subtitles into a video using Whisper, running entirely in your browser — nothing is uploaded, and there's no per-minute charge.
Auto-caption a video
Open Add Captions to Video