Camera Text Reader — OCR That Reads Aloud

While the camera is on, a tone plays: the better the text fills the frame, the higher and steadier it gets. When the frame is caught and stable, the photo is taken automatically and a voice says “Captured”. So you can photograph text without seeing the screen.

More than twenty, including Latin, Cyrillic, Greek, Arabic, Hebrew, Devanagari, Thai, Korean, Japanese and Chinese. Auto mode picks the language from the picture, and it is reliable when the page is in one language and unreliable when two share a line — a Russian label with an English brand name on it is the classic case, and setting the language by hand fixes it. What limits accuracy far more than the language is the shot: even lighting, the page filling the frame, and the camera parallel to the paper. A photograph taken at an angle turns straight lines into curves, and recognition of a curved line of text fails in every script equally.

Accuracy depends on lighting, focus, font and angle. The tool auto-straightens, removes shadows and fixes rotation, but for best results shoot in good light, hold the camera steady and closer to the text. Handwriting is recognized poorly.

Almost everything that matters here, and mostly in the phone's favour. The rear camera of a phone focuses close and moves with your hand, which is the natural way to read a label or a menu; a laptop camera is fixed, faces you, and cannot be pointed at a page. Flashlight and zoom are available where the device offers them, which in practice means Android more often than iPhone. Reading aloud uses the voices installed in your own system, so the same text sounds different on iOS and Android — and on a phone with no voice for that language, the text appears but is not spoken.

Everything except the very first load and the natural HD voice: that one is synthesised on our server, so offline the device voice reads instead. The recognition engine, the language data and the voices are fetched once and then kept by the browser, so after you have used a language once, pointing the camera at a page and hearing it read back works with no connection at all — which matters, because the moments you most need to read a label are often the moments with no signal. Two caveats. A language you have never used has not been downloaded yet, so pick up the ones you need while you have a connection. And clearing site data throws the cache away, which means the next use will need the network again.