Capture: OCR and speech recognition
Origa can read Japanese from images and transcribe it from audio. Both features run on your device — the processing never sends your photos or recordings to a server. This page covers how to use them and what their limits are.
Optical character recognition (OCR)
OCR extracts Japanese text from images. Use it when you have a photo of a menu, a page, a sign, a screenshot, or any other visual source of Japanese text.
How to use:
- Open the words screen and choose the Image tab.
- Drop or paste an image, or pick a file from your device.
- Origa runs OCR and shows the recognized text.
- Review the extracted words, pick the ones you want, and add them as cards.
Supported formats: PNG, JPEG, WebP. A maximum file size applies.
When the model downloads: The OCR model is not bundled with the installer. It downloads the first time you use the feature, with a progress indicator. After the first download, OCR works offline.
What OCR handles well: printed text, clean screenshots, typed Japanese.
Where OCR struggles: heavily stylized fonts, low-contrast photos, vertical text set in unusual layouts, very small characters. Handwritten text is not supported.
Speech recognition (STT)
STT transcribes spoken Japanese from an audio file. Use it when you have a recording of speech — a podcast clip, a line from a show, a voice memo.
How to use:
- Open the words screen and choose the Audio tab.
- Upload an audio file.
- Origa transcribes it on-device and shows the recognized text.
- Review and add the words you want.
Supported format: WAV only. Other common formats (MP3, M4A, OGG) are not currently accepted — convert your file before uploading.
When the model downloads: As with OCR, the STT model is not bundled. It downloads on first use, then runs offline.
What STT handles well: clear single-speaker audio, slow to moderate pace.
Where STT struggles: overlapping speakers, heavy background noise, very fast speech, strong accents. Recognition is approximate; review the output before adding words.
Privacy
Both OCR and STT run entirely on your device. The image or audio you supply is processed locally and is never uploaded. The only network traffic is the one-time model download from Origa's CDN.
When to use capture
Capture is not the primary way to add vocabulary — typing is faster for words you already know. Capture shines when:
- You are reading physical material (a book, a sign) and want to capture unknown words without typing.
- You have a screenshot from a manga reader or a subtitle and want to turn it into cards.
- You heard a word in audio content and want to look it up without typing what you (approximately) heard.