Fix Czech language diacritics #13914
Replies: 4 comments 7 replies
-
|
How did you test Czech lang? Can you provide a demo? |
Beta Was this translation helpful? Give feedback.
-
|
I followed a guide from YouTube and I did exactly the same process, like with the Chinese/English bilingual subs. I tested with two versions of the video, one of them with yellow subtitles on solid black background, another with white subtitles on transparent background, the results pretty much the same. Most of special characters coming out with the "umlaut" mark, some of them with regular accent (long vowel) mark, but no small hooks, etc... |
Beta Was this translation helpful? Give feedback.
-
|
Actually, it looks like most of the hooks excerpt for "ě" were just ignored, and it is mostly the little accents and little circles, which came out as umlauts... 3 4 ... 00:01:27,483 --> 00:01:32,151 11 12 13 |
Beta Was this translation helpful? Give feedback.
-
|
In detail, I found: |
Beta Was this translation helpful? Give feedback.
-
|
The Czech video was a short one, three minutes, 24 subtitle lines. |
Beta Was this translation helpful? Give feedback.
-
|
The subtitles are very clear |
Beta Was this translation helpful? Give feedback.
-
|
Doba ledovä byla ideälni |
Beta Was this translation helpful? Give feedback.
-
|
Should be: Doba ledová byla ideální pro zimní sporty. |
Beta Was this translation helpful? Give feedback.
-
|
The Czech language is treated as a Latin language in processing, using the character dictionary from https://github.com/PaddlePaddle/PaddleOCR/blob/main/ppocr/utils/dict/latin_dict.txt. Check if it includes all Czech characters. If the dictionary is appropriate, you might need to fine-tune the OCR models for your specific scenario. |
Beta Was this translation helpful? Give feedback.
-
|
Oh, It seems that some of these characters are missing (not all of them), indeed. How to fix that? Where exactly to add them? |
Beta Was this translation helpful? Give feedback.
-
|
You need to complete the characters and retrain the model. For specific steps, please refer to https://paddlepaddle.github.io/PaddleOCR/en/ppocr/model_train/recognition.html#12-dictionary |
Beta Was this translation helpful? Give feedback.
-
Hi, this is an awesome project. I first tested with Chinese/English subtitles and was quite impressed. However, testing the Czech language, I found a weird thing - it comes out quite well, except for one significant detail: all diacritical marks come out with an umlaut like in German ü ö, etc., and no correct diacritics for Czech letters like ě,š,č,ř,ž.... Could anyone fix this, please?
Beta Was this translation helpful? Give feedback.
All reactions