Subtitle Edit

[Yosho]

New bug happening after AI updates.
AI transcriptions are now causing all lines to be transcribed and not only the line highlighted.
When transcription is done, it resets back to the first line in the subtitle.

[markfilipak]

This is going to be a long report. It includes many 'creature comfort' requests to make SE easier & quicker to use, and it includes what I think are a few bugs. I hope it pleases.

First, Thank you for sharing SE. It's perhaps the most complex, most difficult-to-write application I've ever seen. I consider its creation to be courageous indeed. I commend all the writers. They care, and it shows. I appreciate that there are actual buttons that push in. That makes it easy to see I've actually done something.

Overall: There is source view and list view. I suggest they be unified to make a single view having subtitle items on single lines, like list view now has. I suggest that the lower edit boxes be removed and that editing be enabled on the single lines that are above the current edit boxes. If the subtitle text is wider than the SE window, then wrap it within the subtitle text column; otherwise, don't.

Desired -- asterisks denote what would be the biggest time savers:
- Could there be an overtype mode? The [Insert] key would be used to toggle between insert and overtype.
- If there is highlighted text when [F3] is pressed, could 'Find' be directly executed on that highlighted text without having to open a 'Find' dialog.
* Could the text scroll up and down so that the current subtitle is always vertically centered?
* When editing times, I have to slow way down -- position mouse, click, long hold, swipe, click-release, type replacement time. That long hold must be there to prevent selecting HH:MM:SS:mmm as a whole; it breaks my rhythm. Perhaps double-click could select HH or MM or SS or mmm while [Shift]double-click would select HH:MM:SS:mmm as a whole. That would be very nice and would just about double my edit speed.
* Could [down-arrow] always position the cursor at the beginning of the next line?
* Could [up-arrow] always position the cursor at the beginning of the previous line?
* Could punctation be treated as white space when selecting entire words? In that regard, I've often dreamed of the following selection functionality: double-click selects entire word (but not any trailing punctuation). Click inside the existing selected string extends the selection to the end of the next word. Click inside the the existing selection extends it to the end of the next word. Et cetera. I've never seen an editor that would do that. It would be heavenly.
- Lacking vertical centering, could the cursor line be hightlighted so that it's easy to find. A pale background color is preferred. Highlighting across the entire width of the pane would be fine and might be simpler.
- When switching back and forth between list & source views, could they 'inherit' index numbers and cursor positions, one from the other. [note1]
- Could SE show that a video file has already been loaded, and show its name.
- Could SE tag particular entries for future reference, so that users can 'jump' from tag to tag like 'jumpping' forward and backward between 'finds'.
[note 1] Switching back and forth is required because list & source views have differing capabilities. I find list view of little use except as a means to add/delete/split/merge entries and to select all entries in order to [Auto br] all entries at once.

Undo: [Ctrl][Z] has major cursor positioning problems. After [Ctrl][Z], the cursor is in the wrong place. It does not jump to the undo point, so I can't tell what's been undone. Could you arrange it so that [Ctrl][Z] does not cause the current line to jump to the bottom of the source window? Could a single [Ctrl][Z] undo all the piecewise changes that have been made to the current entry and then stop? Currently, I can't track what's been undone and I get lost. When I do multiple [Ctrl][Z]s, I can't tell what's been undone. Likewise, when I do multiple [Ctrl][Y]s, I can't tell what's been redone. I think the easiest way to handle cursor position is to move the cursor (and vertical center) to the cursor position it had _before_ the undone text was first inserted, and to move the cursor (and vertical center) to the cursor position it had _after_ the redone text is re-inserted. That will probably require recoding cursor position with each keystroke in anticipation of possible undos.

Notifications: Notifications that cite line numbers are useless. The subtitle instances are numbered by instance, not by line. Any particular line number can't be found.

Initialization: If I forget to toggle source-/list-view a few times before doing my first 'Find', 'Find' doesn't work and shows 'not found' in every search.

'Multiple replace...': If subtitle text is highlighted, 'Multiple replace...' doesn't 'see' the highlighted text. For example, see
'SE Multiple replace... while text is selected bug.png'.
In that example, quitting 'Multiple replace...', unhighlighting the word "confine", and repeating 'Multiple replace...' works as expected. I have repeated the test with other words.

[Change all]: [Change all] currently doesn't change all. For example, a character's name, "Troi", appears as "Trol" over and over in the same session despite [Change all] correction. That experience is repeated in every session. [Change all] seems to always work on words that are not "Trol". I don't know what it is about the word "Troi" that is causing this malfunction.

Cursor position: For selected text -- selected via 'Find' for example -- have [<] put the cursor before the selection's beginning character and [>] put the cursor after its ending character. Currently, if "3456" is selected within "12345678", [<] puts the cursor here:
12345|678, and [>] puts the cursor here:
1234567|8.
To be consistent with standard text editors, [<] should put the cursor here:
12|345678, and [>] should put the cursor here:
123456|78.

[Replace all]: [Replace all] replaces only from the current subtitle to the end. It does not replace all.

Exception:
'Windows Forms Thread Exception.png'
show an exception during an attempted regexp Find. The exception is thrown every time I attempt a regexp Find.

Lack of focus: Switching from source view to list view, no list view element has focus, so [up-arrow]/[down-arrow] doesn't function and [Ctrl][A] selects nothing.

[markfilipak]

I see there are 'shot changes'.
I see this control: 'Tools'>'Adjust durations...' : '[X] Don't extend past shot changes'
I see this storage: c:\Users\Administrator\AppData\Roaming\Subtitle Edit\ShotChanges\

How do I make a ShotChanges table for use by 'Adjust durations...'? I have not found such a tool.

Thanks--Mark.

[Emulgator]

You may just have SE generate one and handedit afterwards. It is seconds.milliseconds format, ANSI coding.
ssss.ccc
The filename seems to be generated unique and random from 16 hex digits,
so I do not know how to push that into a project, I just let SE generate one, see above.
Invoking a rescan seems not implemented ATM, but I guess if you just rename the one belonging to the project (compare by modification date) a rescan should become available again.
BTW I decided to choose 0.2 as sensitivity, larger or default and some more shotchanges are missed.

[markfilipak]

Originally Posted by Emulgator

You may just have SE generate one ...

How do I get SE to generate one? There doesn't seem to be a menu function to do that.

[markfilipak]

Let me explain.

In the last week I've modified the subtitles of 140 episodes of STAR TREK, THE NEXT GENERATION. It was tedious but went smoothly. The subtitles for the first 5 seasons were quite sloppy; they always began at the right times but often slopped over into the next scene by as much as 10 seconds. I had to manually edit the ending sexagesimal times.

Season 6 is different. The subtitle times span the vocalizations and that is too short. I want them to start at the given times but with 125% durations. Further, I want them clipped to prevent slopping into the next shot. To that end, I've loaded the MP4. However, I can't find a way to scan the MP4 to make a shot change table.

Also, I've not found 'Help' that still exists.

[Emulgator]

Just press the button "Beautify time codes" (red/green waveform)
(If this button is not visible in Toolbar: Settings -> Toolbar -> toggle "Beautify time codes" visible)
A window pops up. Tick Use FFprobe... then "Extract time codes".
(You got to download FFprobe the first time, IIRC SE does it for you.)
(This button can be activated only the first time, is then greyed out as long as the .timecodes file is present I guess)
Parsing might take considerable time (15..25min), depending on duration, codec and disk speed.
After successful parsing you will be presented with the message "128568 time codes loaded"
Then "Generate shotchanges". I suggest to lower sensitivity to 0.2 to catch them all.
(This button can be activated only the first time, is then greyed out as long as the .shotchanges file is present I guess)
Again Parsing might take considerable time (15..25min), depending on duration, codec and disk speed.
After successfull parsing you will be presented with the message "398 shot changes loaded".

If you prefer to have all audio languages muxed in to cater for any timing of any language (as I do):
It does matter from which container you decide to work from when attempting to switch audio streams and still be able to see waveforms.
I prepare a SE Mux just for that occasion.
.mkv seems to work well, .mp4 seems to work well.
.m2ts may have the BD-typical muxing delay which will shift any timestamps and hinder work.

P.S. Just checked it, I had renamed my last project's shotchange file again and again,
had SE parse again and again until I was happy with the result.
Fired up WinMerge to see the 3 versions side-by-side and yes, they filled up gradually.
Ah, and the .timecodes file in \TimeCodes folder has the same unique and random filename as the accompanying .shotchanges file.

So: Happy editing.

[markfilipak]

Originally Posted by Emulgator

Just press the button "Beautify time codes" (red/green waveform)
(If this button is not visible in Toolbar: Settings -> Toolbar -> toggle "Beautify time codes" visible)
A window pops up. Tick Use FFprobe... then "Extract time codes".

Ah! "Beautify time codes"! That's where it's buried. Thank you. ...odd place to put such a vital function, eh?

... After successful parsing you will be presented with the message "128568 time codes loaded"
Then "Generate shotchanges". I suggest to lower sensitivity to 0.2 to catch them all.

Well, If I was writing the code to detect shot changes, I'd pick frames having few frame-to-frame pixel correlations plus no/few motion vectors, but I imagine simply few pixel correlations is what is done. I also reckon that 0.2 is actually higher sensitivity -- a seemingly inverse metric that's inverse because it's based on percentage of correlation. I reckon running shotchanges takes so long because FFmpeg is decoding. Just looking at motion vectors would not require decoding and would be faster; maybe significantly faster. But of course lack of motion vectors is also characteristic of still images, so both MVs and pixel correlations would be needed to accurately assay shot changes.

So: Happy editing.

You're extraordinarily helpful and complete. I hope you live as long as you want, in perfect health, and blissfully happy.

--Mark.

[Emulgator]

Many thanks, Sir, for your kind words, and may I return the favor.

[markfilipak]

I recently discovered that 'Edit image db' > 'Add better match', is a very effective way to accelerate/improve training. Unfortunately, I haven't found a way to get a subtitle line that has a bogus '2'-to-'?', for example, into 'Edit image db' because 'Edit image db' is not available as a general tool.

nikse, I apologize if this is not the right place to make this request. ...I'm so impressed with SE!

--Mark.

PS: nikse, I want to make a large donation, however, I will not use PayPal. There is a better way. Kindly contact me privately. --M.

[markfilipak]

Comments, please...

I've learned to adjust timing solely via 'Synchonization'>'Adjust all times'>'(o) Selected and subsequent lines'. I tried to do 'Point sync' of a region but it didn't work out.

The region was 13:50.413..22:31.267. I specified ends that have zero deltas, but between them the times delta by nearly a second, some leading, some lagging.

Since the ends had zero deltas, I assumed the deltas I specified would affect only the times between those ends. I was disappointed to discover that _all_ the subtitle times were changed. Specifically, the initial subtitle was moved backwards by about 13 seconds and now appeared less than a second into the movie.

Perhaps there's a way to limit 'Point sync' to a certain set of subtitles, but I didn't see any way to do that.