Doom9's Forum - View Single Post

ai4spam · 24th May 2005, 05:23

@Esc:
1) Will take care of this.
2) This happens to me with diactitics in languages like Romanian. It's not SubRip's fault, somehow you changed your system settings (that, plus Delphi makes a hidden ANSI-OEM-ANSI conversion in strings, and it can't be taken out AFAIK). Just go to regional settings and set "language of non-unicode programs" to the language of your choice. Are you sure you're using the é from the default font/charset? In my example (Romanian), some letters would change (t, and s,), but not others (i^ and a^). The way I "fixed" it was to assign the correct characters to some of the 20 buttons on the bottom. If you select the Romanian language in the OCR window, you'll notice that the letters I mentioned still don't show up correctly, even when using the EASTEUROPE charset, but they do show up correctly in the final subs. So, try to play with the charset in the general options window. The rule for characters like your é is: if you copy and paste it on a button (with right click), it should show up as you want it on the button, or else it's not the right character.
3) I fixed it, will upload in next beta.

@Longinus:
Thanks for your appreciation.
Your "transparent window" bug is really weird, there is no reason why it should happen. Maybe there's a problem with your video drivers? Please try it on another machine and let me know if it still happens. A screenshot would also be useful. Ah, and it's worth asking: are you sure you can't see the text because of the new feature (fill to the sides of the text with fuchsia color)?
The skip first, update every and min duration were set there for temporal optimization, without them it would be really slow. Beats me why you would try to OCR a timecode...
Anyway, basically I'm processing every frame, when I detect a sub I skip the first frames (to let the sub appear fully), then accumulate for min duration frames, then just compare and reset every now and then.
In your case, if the timecode is the only thing you OCR, then it doesn't change much from frame to frame. Try also lowering the same sub tolerance to a really small number. However, it will still accumulate at least 1 frame, to process every frame I'll need to make some changes (again, done, but will upload in next beta).

24th May 2005, 05:23	#79 \| Link
ai4spam Programmer Join Date: Sep 2003 Posts: 382	@Esc: 1) Will take care of this. 2) This happens to me with diactitics in languages like Romanian. It's not SubRip's fault, somehow you changed your system settings (that, plus Delphi makes a hidden ANSI-OEM-ANSI conversion in strings, and it can't be taken out AFAIK). Just go to regional settings and set "language of non-unicode programs" to the language of your choice. Are you sure you're using the é from the default font/charset? In my example (Romanian), some letters would change (t, and s,), but not others (i^ and a^). The way I "fixed" it was to assign the correct characters to some of the 20 buttons on the bottom. If you select the Romanian language in the OCR window, you'll notice that the letters I mentioned still don't show up correctly, even when using the EASTEUROPE charset, but they do show up correctly in the final subs. So, try to play with the charset in the general options window. The rule for characters like your é is: if you copy and paste it on a button (with right click), it should show up as you want it on the button, or else it's not the right character. 3) I fixed it, will upload in next beta. @Longinus: Thanks for your appreciation. Your "transparent window" bug is really weird, there is no reason why it should happen. Maybe there's a problem with your video drivers? Please try it on another machine and let me know if it still happens. A screenshot would also be useful. Ah, and it's worth asking: are you sure you can't see the text because of the new feature (fill to the sides of the text with fuchsia color)? The skip first, update every and min duration were set there for temporal optimization, without them it would be really slow. Beats me why you would try to OCR a timecode... Anyway, basically I'm processing every frame, when I detect a sub I skip the first frames (to let the sub appear fully), then accumulate for min duration frames, then just compare and reset every now and then. In your case, if the timecode is the only thing you OCR, then it doesn't change much from frame to frame. Try also lowering the same sub tolerance to a really small number. However, it will still accumulate at least 1 frame, to process every frame I'll need to make some changes (again, done, but will upload in next beta). Last edited by ai4spam; 24th May 2005 at 05:35.