Language

[SangenBR]

I would like to ask for support for Portuguese and Brazilian Portuguese, is it possible?

[arition]

I am happy to support Portuguese and Brazilian Portuguese, but I know little about these languages. If you can provide me some sample Portuguese subtitle files (ass files, more is better), I can train a model for you.

[SangenBR]

send subtitles in your email
do you train hardsub video subtitles?

[arition]

Received. Training takes time, so please be patient.
For details about training, please check https://github.com/freyjaSubOCR/freyja-sub-ocr-training.

[SangenBR]

do in your time without haste, thanks

[arition]

Progress report

Training for Portuguese (or any kinds of Latin languages) is much harder than I think. The main problem is that sentences in Latin languages are much longer than the sentences in Chinese, so the model structures need to be changed. However I am really busy recently and I do not have time to try different kinds of network structure and parameters. The variable width of Latin characters also hurt the performance of the model.

Good news is that I managed to train a baseline model. The accuracy is not very high, but it is usable. I will try to update the GUI and provide a test version for you if I have time this week.

[SangenBR]

Oh, I would like to test, but do it in your time I'm not in a hurry.

[SangenBR]

I see that your problem is the opposite of mine because I think that the Chinese language has few characters and reading and very fast it gets to me when I see some Chinese anime