![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
For almost four years, I've had a plain txt file on my desktop where I kept sentences and such I intended to enter into Mnemosyne for Japanese study later. Sentences went in when I read stuff. Sentences came out. It got longer and it got shorter. But the file was never entirely empty.
Until today. I reached bun.txt zero and just deleted it :D
I'd also like to share a tool I found recently called Capture2Text, which is essentially a desktop OCR application that can (try to) read from anything on your screen - very helpful for image sources likes comics that you can't copy from. It can handle vertical text for Asian languages and has add-ons for a lot of languages, from Arabic to Tibetan to math equations.
My 2.5 seconds of experimenting with the English setting found excellent results. The Japanese is spottier - sometimes it captures perfectly with no trouble, and others it's like, え = 乙, right? But it read surprisingly well from pictures I took from a textbook that I tried, and it works almost perfectly on fan comics I've used it on, too; in both cases, it's faster than processing the image through Google or manually typing it up (esp. if there's a kanji I can't identify or remember the reading of).
I'm still experimenting to see how to get it to work best - so far it's has more trouble on an older manga I purchased in digital form vs doing so well on Twitter comics. Some experiments in Photoshop lead me to think noise (it's not a very clean release and there's lots of jpg artifacts) and furigana can trip it up; the difference in font choice may affect the results, too. It also sometimes works better on smaller text, as long as the kanji are still clear.
Anyway, despite its imperfections, it's already proven useful to me, so I thought I'd throw the link out for anyone else who hasn't heard of it.
Until today. I reached bun.txt zero and just deleted it :D
I'd also like to share a tool I found recently called Capture2Text, which is essentially a desktop OCR application that can (try to) read from anything on your screen - very helpful for image sources likes comics that you can't copy from. It can handle vertical text for Asian languages and has add-ons for a lot of languages, from Arabic to Tibetan to math equations.
My 2.5 seconds of experimenting with the English setting found excellent results. The Japanese is spottier - sometimes it captures perfectly with no trouble, and others it's like, え = 乙, right? But it read surprisingly well from pictures I took from a textbook that I tried, and it works almost perfectly on fan comics I've used it on, too; in both cases, it's faster than processing the image through Google or manually typing it up (esp. if there's a kanji I can't identify or remember the reading of).
I'm still experimenting to see how to get it to work best - so far it's has more trouble on an older manga I purchased in digital form vs doing so well on Twitter comics. Some experiments in Photoshop lead me to think noise (it's not a very clean release and there's lots of jpg artifacts) and furigana can trip it up; the difference in font choice may affect the results, too. It also sometimes works better on smaller text, as long as the kanji are still clear.
Anyway, despite its imperfections, it's already proven useful to me, so I thought I'd throw the link out for anyone else who hasn't heard of it.
