Subtitles (closed caption) are in the wrong order compared to the audio

maineiac13

Downloading Gilmore Girls Season 1, episode 2 from Netflix (and also other episodes). In many cases the closed caption subtitles are in the reverse order of the audio. Examples: At about the 20 second mark the audio is: "And bad girls always wear red nail polish." But the subitles read:

ALWAYS WEAR RED NAIL POLISH.
AND BAD GIRLS

At the 39 second mark the audio is: "Who Cares! You're gonna be wearing shoes anyway!" But the subtitles read:

GONNA BE WEARING SHOES ANYWAY!
Lorelai: WHO CARES?! YOU'RE

I tried this with both imbedded subtitles and subtitles put into a separate SRT file...with the same result.

What is also interesting is that this reversal of subtitle text does not occur in all cases where there are two lines of subtitle text, just in many (not sure if most or if a minority). Makes it very difficult at times to follow the conversation.

This happens in most if not all of the episodes
Using 6.2.5.5 Windows x64 version

Attached is my log file: StreamFab.log

Germania

At the 39 second mark the audio is: "Who Cares! You're gonna be wearing shoes anyway!" But the subtitles read:

GONNA BE WEARING SHOES ANYWAY!
Lorelai: WHO CARES?! YOU'RE
maineiac13

My default for NF+PV is "Extract Original Format" and this is correct:

Germania

I can confirm that the problem occurs with all VIP Video Formats and remux into file

maineiac13

Yes, if I watch the episode directly on Netflix, the subtitles appear in the proper order, but when they are downloaded that is when the problem occurs. So clearly the problem is with something that occurs when SF is downloading/creating the subtitles.

Stan001

Netflix for me always had issues with srt. I only did remux, but I do see that no longer works right either, at least not for this show.

Only 2 choices do what Germania states and select Extract Original Format, and it saves as a vtt file. And then works fine.
Or provide feedback directly to SF and let them know, as if it can be fixed, only the developer can fix it.

maineiac13

Well, I tried the "Extract Original Format" method but that did not work. Not sure why it is working for Germania and not for me:
(Note that the vtt file for this subtitle reads as follows:

26
00:00:38.956 --> 00:00:42.001 position:19.52%,start align:start size:67.14% line:84.62%
GONNA BE WEARING SHOES ANYWAY!

27
00:00:38.956 --> 00:00:42.001 position:21.43%,start align:start size:61.43% line:79.29%
Lorelai: WHO CARES?! YOU'RE

which would explain the placement on the video, but based on the audio the two lines should be reversed, and they are correct if I just watch the video directly on Netflix without downloading.

The image below shows what I see when watching after downloading using the "Extract Original Format" method

Germania

Not sure why it is working for Germania and not for me:
(Note that the vtt file for this subtitle reads as follows:

26
00:00:38.956 --> 00:00:42.001 position:19.52%,start align:start size:67.14% line:84.62%
GONNA BE WEARING SHOES ANYWAY!
maineiac13

I converted original vtt with (portable) SubtitleEdit to srt
via context menu: SubtitleEdit.exe /convert "%1" srt

Both lines have same time stamp ..
.. but also vlc is showing correct:

jpp72

I use faster-whisper to generate English subtitles for Japanese videos that don't have them, and I get those issues often.

In my script, i use


SubtitleEdit.exe /convert "%basename%.en.srt" srt /fixcommonerrors /overwrite

On all AI-generated subtitles and that command fixes those issues.

Maybe worth a try...

maineiac13

I converted original vtt with (portable) SubtitleEdit to srt
via context menu: SubtitleEdit.exe /convert "%1" srt

Both lines have same time stamp ..
.. but also vlc is showing correct:

Germania

I don't convert to srt because I generally watch these videos with Kodi which (unlike a number of others) readily reads vtt.

Yes, all the problems with these subtitles are when multiple text lines have the same time stamp, but the first one is the one that should be the last. It is interesting that in your case, the subtitles got put back together properly even though the wrong one is listed first. Wonder if that is the case with the other one I mentioned at about the 20 second mark. Also, curious if you did not convert to srt would the subtitles be in the right order. In my case using the vtt file directly the text is in the wrong order.

maineiac13

I use faster-whisper to generate English subtitles for Japanese videos that don't have them, and I get those issues often.

In my script, i use
SubtitleEdit.exe /convert "%basename%.en.srt" srt /fixcommonerrors /overwrite
On all AI-generated subtitles and that command fixes those issues.

Maybe worth a try...
jpp72

I also occasionally use a subtitle generator but really don't want to do so when I am already paying for a service that should be able to simply download the subtitles properly. Moreover, these are closed captions which often include words that come from sounds, signs, etc not just the spoken words and thus would not generally be captured by a subtitle generator.

Another thing I forgot to mention was that originally I had downloaded the episodes using 6.2.5.3. Those downloads were much worse than the 6.2.5.5 downloads. In the former almost every 2-line subtitle throughout the episode was reversed. With 6.2.5.5 the problem occurs much less frequently so SF must have done something which improved the situation, though the developer may not have been actually directly addressing this issue???

Germania

I don't convert to srt because I generally watch these videos with Kodi which (unlike a number of others) readily reads vtt.
maineiac13

Original vtt (and converted to srt) works fine with mpc-hc (my default)
and of course the same with mpc-be - also successfully tested with vlc.

So it must be a Kodi problem.

jpp72

I also occasionally use a subtitle generator but really don't want to do so when I am already paying for a service that should be able to simply download the subtitles properly. Moreover, these are closed captions which often include words that come from sounds, signs, etc not just the spoken words and thus would not generally be captured by a subtitle generator.
maineiac13

I meant running SubtitleEdit on the subtitles you downloaded to fix the problems.

jpp72

Original vtt (and converted to srt) works fine with mpc-hc (my default)
and of course the same with mpc-be - also successfully tested with vlc.

So it must be a Kodi problem.
Germania

I use Kodi for EVERYTHING i play. Even when watching on my phone with my Viture Luma Pro (which are amazing on a long plane trip BTW)

Maybe has something to do with the fact that i strip all formatting and positioning from all subtitles...

Germania

Maybe has something to do with the fact that i strip all formatting and positioning from all subtitles...
jpp72

I can't say how Kodi behaves when two subtitle texts have the same time stamps ..

.. the fact is that all of my installed players display it correctly.

october262

I also occasionally use a subtitle generator but really don't want to do so when I am already paying for a service that should be able to simply download the subtitles properly. Moreover, these are closed captions which often include words that come from sounds, signs, etc not just the spoken words and thus would not generally be captured by a subtitle generator.

Another thing I forgot to mention was that originally I had downloaded the episodes using 6.2.5.3. Those downloads were much worse than the 6.2.5.5 downloads. In the former almost every 2-line subtitle throughout the episode was reversed. With 6.2.5.5 the problem occurs much less frequently so SF must have done something which improved the situation, though the developer may not have been actually directly addressing this issue???
maineiac13

do you have "Hulu" ?? if yes, then try downloading Gilmore Girls Season 1, episode 2 with your subtitles and see if you get the same results.
https://www.imdb.com/title/tt0238784/?ref_=nv_sr_srsg_0_tt_8_nm_0_in_0_q_Gilmore%2520Girls%2520

october262

I also occasionally use a subtitle generator but really don't want to do so when I am already paying for a service that should be able to simply download the subtitles properly. Moreover, these are closed captions which often include words that come from sounds, signs, etc not just the spoken words and thus would not generally be captured by a subtitle generator.

Another thing I forgot to mention was that originally I had downloaded the episodes using 6.2.5.3. Those downloads were much worse than the 6.2.5.5 downloads. In the former almost every 2-line subtitle throughout the episode was reversed. With 6.2.5.5 the problem occurs much less frequently so SF must have done something which improved the situation, though the developer may not have been actually directly addressing this issue???
maineiac13

also see - https://kodi.wiki/view/Subtitles

maineiac13

I think I have (maybe) figured out what is happening.

First, as Germania noted the problem occurs with Remux Into File. It also occurs with "Extract to SRT file". However, depending on the video player, the problem may or may not show up when using "Extract Original Format" which creates a ttf file. Players such as vlc play the lines correctly. Players such as Kodi and Plex do not. But why? I think it may relate to how those players read the lines in question. Looking at lines 26 and 27 in the ttf file they read as follows:

26
00:00:38.956 --> 00:00:42.001 position:19.52%,start align:start size:67.14% line:84.62%
GONNA BE WEARING SHOES ANYWAY!

27
00:00:38.956 --> 00:00:42.001 position:21.43%,start align:start size:61.43% line:79.29%
Lorelai: WHO CARES?! YOU'RE

Line 27 should be the top line in the video, with line 26 the bottom line. My guess is that Kodi and Plex and others simply read the lines in numerical order when the time frames are the same so that the top line shown on the video is the text in line 26 and the bottom line is the text in line 27, thus the language is reversed from the audio.

But vlc and others read those lines in the opposite order, placing the text of line 27 on top of the text in line 26. Why? My only guess is that they not only read the time frames but also take into consideration the line percentage which would have the text in line 27 be on top of the text in line 26.

Thoughts?

maineiac13

I use Kodi for EVERYTHING i play. Even when watching on my phone with my Viture Luma Pro (which are amazing on a long plane trip BTW)

Maybe has something to do with the fact that i strip all formatting and positioning from all subtitles...

jpp72

I have tried various settings for the formatting and positioning of subtitles in Kodi but without any improvement. In fact, in some cases the situation actually gets worse as many more cases of the reversed subtitles show up. I think I will try to do some more work on this and possibly go to the Kodi forums and see what the cause may be and if there is a solution.