I made a post before explaining my philosophy about learning via immersion and that the best way to advance beyond N3-N2 is to read a lot of native material.
There’s one problem: when you encounter a Kanji you’re not familiar with it can be duanting to look it up. There are various ways: look it up by stroke order, by components, by an OCR app, etc. But it always takes time.
I also mentioned in that thread that I was working on something to help with this.
So I made a utility to help with this. I’m calling it “Yomitai” because it’s what I want to use when I want to read.
It’s based on OCR, but instead of trying to point the camera at the right character and freezing the screen, you just take a picture of the whole page (or a screenshot, if it’s an ebook with DRM that doesn’t let you select the text, or if the ebook is an image scan to begin with).
Then you can point at the characters you don’t know one by one to see the reading and meaning.
I’m putting out for public beta testing. Expect it to have some issues, but in my testing it seems to work equally well with screenshots and real life pictures of physical books.
https://yomitai.app/
You can see some demo videos on the homepage.
When you point at the word it not only shows the reading and meaning, it also pins it on the side.
Basically I use the app not only to OCR the kanji but as the reading medium, with color-coded annotations to help look up the same word again visually.
Current known limitations:
– Does not work on mobile phones
– Does not persist your work
– Some problems when you try it on manga
The OCR is based on Google’s vision APIs, which uses machine learning, so the quality of the detection is much higher than what you would get with some apps that use Tesseract (I tried it, didn’t work so well).
by hasen-judi