11 September 2013 Karim Elatov

I recently had a situation where I had 2 Avi files and 2 Sub/Idx files, and I just wanted to combine them into one Avi file and one srt file. Here are the files that I started out with:

$ ls
file1.avi  file1.sub  file2.idx
file1.idx  file2.avi  file2.sub

Difference between VobSub and SRT

From Subtitle Formats Explained:

Most subtitles consist purely of text characters. Since text is also some of the easiest data to store and compress it makes sense to store subtitles as simple text files or a text stream within a video file. Although it’s normal for all subtitles to start out this way, that doesn’t mean that’s how they’re stored.

As a matter of fact subtitles on DVDs aren’t actually text. They’re actually encoded as raster graphics. Much like the way characters on older text-based computer interfaces, they’re actually just a collection of dots on a grid. These images are put over the top of the video frame when displayed.

More from the same site:

VobSub subtitles have become very common because it’s easy to get them from DVDs. In fact, VobSub basically just re-packages the images from the DVD into a file that has the extension of .SUB and additional information in another file with an extension of .IDX. They’re generally referred to as either VobSub or IDX + SUB (IDX/SUB) subtitles. Information in the IDX file tells media player software the color of the subtitles, their position on the screen, when they appear and disappear, and a number of other important pieces of information.

So VobSub format consists of images and a descriptor file to correlate the times when each image should be displayed (this is a very oversimplified definition … there are a lot of other aspect to it as well). Here is similar description from VobSub:

VOBsub extracts the DVD subtitles raw PES from a DVD and dumps this to a .sub file. It also creates a .idx Index file with the times and byteoffsets for each and every single subtitle. The format has support for multiple tracks and can also be embedded in MP4 and Matroska files.

For SRT or (SubRip text file format), we can check out the wikipedia page:

The SubRip file format, as reported on the Matroska multimedia container format website, is “perhaps the most basic of all subtitle formats.”SubRip (SubRip Text) files are named with the extension .srt, and contain formatted lines of plain text in groups separated by a blank line. Subtitles are numbered sequentially, starting at 1. The timecode format used is hours:minutes:seconds,milliseconds with time units fixed to two zero padded digits and fractions fixed to three zero padded digits (00:00:00,000).

There is a lot more information regarding SubRip available here:

So SubRip Format is a text file with the subtitles and timing included (again, overly simplified). There are actually a bunch of other formats. Here is a table from the wikipedia page:

different format of subtitles Combine VobSub (sub/idx) Format Subtitles into a Single SubRip (srt) Format File

But I will just be working with vobsub and srt.

Merge 2 Avi Files

This part is really easy. You can complete this with either mencoder:

yum install mencoder
mencoder -ovc copy -oac copy file1.avi file2.avi -o files.avi

Or with transcode:

yum install transcode
avimerge -o files.avi -i file1.avi file2.avi

After that you should have one big Avi file.

Convert VobSub (SUB/IDX) files into SRT subtitles with subtitleripper

This is actually the difficult part.

Check for Multiple Subtitle Streams

First we should check if the VobSub files have multiple languages. You can do this with either mediainfo:

$mediainfo file1.sub
General
Complete name                            : file1.sub
Format                                   : MPEG-PS
File size                                : 3.80 MiB
Duration                                 : 59mn 33s
Overall bit rate                         : 8 921 bps

Text #1
ID                                       : 189 (0xBD)-36 (0x24)
Format                                   : RLE
Format/Info                              : Run-length encoding
Muxing mode                              : DVD-Video
Duration                                 : 59mn 33s

Text #2
ID                                       : 189 (0xBD)-37 (0x25)
Format                                   : RLE
Format/Info                              : Run-length encoding
Muxing mode                              : DVD-Video
Duration                                 : 59mn 33s

Since we have more than two texts that means we have multiple languages in the VobSub files. Notice the ID fields: 0x24 and 0x25. We will actually use those later on. You can also use tcscan utility to check for the subtitle IDs as well:

$ tcscan -i file1.sub
[scan_pes.c] found first packet header at stream offset 0x0
[scan_pes.c] found private_stream_1 stream [0xbd]
[scan_pes.c] found padding stream [0xbe]
[scan_pes.c] end of stream reached
[scan_pes.c] ------------- presentation unit [0] ---------------
[scan_pes.c] stream id [0xbd]   1946
[scan_pes.c] stream id [0xbe]   1572
[scan_pes.c] 3518 packetized elementary stream(s) PES packets found
[scan_pes.c] presentation unit PU [0] contains 0 MPEG video sequence(s)
[scan_pes.c] ---------------------------------------------------
[scan_pes.c] (scan_pes.c) detected a total of 1 presentation unit(s) PU and 0 sequence(s)

Notice again we have two stream IDs: 0xbd and 0xbe which mediainfo converted for us. If you wanted to find out what language those subtitles correlate to, you can actually use mplayer for that. Here is an example:

$ mplayer -vo null -ao null -frames 0 -identify file1.sub -v  2>/dev/null | grep -E '^ID_VOBSUB_ID' -A 1
ID_VOBSUB_ID=4
ID_VSID_4_LANG=en
--
ID_VOBSUB_ID=5
ID_VSID_5_LANG=en

We can see that ID 4 (0x24 or 0xbd) is in english (en) and ID 5 (0x25 or 0xbe) is also in english. There is a brief description from here, regarding the offset and the value being in hexadecimal:

The subtitles in MPEG program stream ( = VOB files) use IDs that start at 0x20. tcextract uses those IDs (in fact, tcextract can extract arbitrary streams from PS files). mplayer simply bases its -sid parameter at 0 and adds 0x20 to that value internally when selecting the appropriate stream.

The last method is by checking the .idx file. Here is what I saw in mine:

$ grep ^id file1.idx -B 1
# English
id: en, index: 4
--
# English
id: en, index: 5

Extract Subtitle into a Raw Stream with transcode

We can use the tcextract utility which comes with the transcode package to extract the raw stream of the desired subtitle. Here is how that command looked like:

$ tcextract -i file1.sub -x ps1 -a $((0x20 + 4)) > file1.ps1

Now you should have a non-zero sized file:

$ ls -lh file1.ps1
-rw-r--r-- 1 elatov elatov 1.2M Aug 29 19:01 file1.ps1

Convert VobSub Subtitle Stream to Images with subtitleripper

Now we can use subtitle2pgm which is from the subtitleripper package to convert the subtitle stream to images. Here is the command for that:

$ subtitle2pgm -i file1.ps1 -o file1 -P
...
...
Generating image: file10786.pgm
Generating image: file10787.pgm
Generating image: file10788.pgm

For every caption it creates an image. I had this many pgm files:

$ ls *.pgm | wc -l
788

You can take a look and see how the image looks like with the following command:

$ feh file10788.pgm

and it will look something like this:

feh after subtittle2pgm Combine VobSub (sub/idx) Format Subtitles into a Single SubRip (srt) Format File

You will also see a .srtx file, this is a description file of the converted images. Here is a snippet from that file:

$ head file1.srtx
1
00:00:59,880 --> 00:01:07,609
file10001.png.txt
2
00:01:07,689 --> 00:01:13,420
file10002.png.txt
3
00:01:13,500 --> 00:01:21,200

With the help of srttool we will be able to create a single .srt file after we have converted the images to text. So first let’s do that.

Convert VobSub Caption Images to Text with tesseract

Let’s install tesseract:

$ yum install tesseract

Here is what I ran to convert the images to text:

$ for f in *.pgm ;  do tesseract "$f" "$f" ; done

When finished, for every .pgm file there was a corresponding .txt file, like so:

$ ls file10788*
file10788.png  file10788.png.txt

Here is the converted text file:

$ cat file10788.pgm.txt
Still, he's got style

Combine All the Text Files into an SRT File with subtitleripper

Since the srtx file (generated from subtitle2pgm) contains the timestamps, we can create an appropriate srt file from all the text files and the corresponding .srtx file. Here is the command to accomplish that:

$ srttool -s -i file1.srtx -o file1.srt

You can confirm the file is okay, by checking out it’s contents:

$ tail file1.srt
787
01:00:25,039 --> 01:00:26,559
What the hell's that?

788
01:00:33,650 --> 01:00:35,510
Still, he's got style

That looks pretty good.

Clean Up OCR Generated SRT File

Since the OCR conversion is not perfect we should check to make sure the text looks good. You can either do this manually, by running this:

$ ispell -d american file1.srt

or this

$ aspell -d en -c file1.srt

Both will ask you to fix words along the way. Here is how it will look like from the terminal:

aspell fix spelling Combine VobSub (sub/idx) Format Subtitles into a Single SubRip (srt) Format File

subtitleripper actually provides a sed file to do this for us. So we can run this:

$ sed -f /usr/share/subtitleripper/gocrfilter_en.sed file1.srt > file1_fixed.srt

You can check all the changes that it made by checking the differences between the files:

$ diff file1.srt file1_fixed.srt
110c110
< ls that right?
---
> Is that right?
3353c3353
< ls his face among
---
> Is his face among

the sed file doesn’t do too much, it just searches for common patterns and replaces them. Here are the contents of that file:

$ cat /usr/share/subtitleripper/gocrfilter_en.sed
# Replace common gocr mistakes in english language
# Please use info sed to obtain more information
# about sed syntax or read
# http://www.ptug.org/sed/sedfaq.htm
# 2002-6-18: Modified to use \<..\> word boundaries.
s/\<l\>/I/g
s/\<l'll\>/I'll/g
s/\<ln\>/In/g
s/\<lt\>/It/g
s/\<lt's\>/It's/g
s/\<l've\>/I've/g
s/\<l'd\>/I'd/g
s/\<-l\>/-I/g
s/\<l?\>/I?/g
s/\<iust\>/just/g
s/\<ls\>/Is/g
s/\.iust /.just/g
s/\<lsrael\>/Israel/g
s/\<lgrayne\>/Igrayne/g
s/\<lf\>/If/g
s/\<l'm\>/I'm/g
s/\<lt'd\>/It'd/g

You can add your own, if think it’s missing something. Finally just rename the files so we can stay organized:

$ mv file1.srt file1_after_ocr.srt
$ mv file1_fixed.srt file1.srt

Convert VobSub (SUB/IDX) files into SRT subtitles with ogmrip

There is a similar tool that does the above in less steps, it’s called ogmrip. First let’s install it:

$ yum install ogmrip

We don’t need the raw stream, we can just convert from VobSub to images directly.

Convert VobSub to Images with ogmrip

Here is the command to do that:

$ subp2pgm file1 -o file1 -s 4
788 files generated

the -s specifies the subtitle ID (so I am picking the first english subtitles). The first file1 specifies the basename for the .sub and .idx files. So you need to have file1.sub and file1.idx in the directory where you run that command from. After that it’s done you will have the 788 pgm images and an xml file (similar to the .srtx file from subtitle2pgm command) for the description file:

$ ls *.pgm | wc -l
788
$ head file1.xml
<?xml version="1.0" encoding="UTF-8"?>
<subtitles>
  <subtitle id="1" start="00:00:59.826" stop="00:01:07.562">
    <image>file10001.pgm</image>
  </subtitle>
  <subtitle id="2" start="00:01:07.634" stop="00:01:13.368">
    <image>file10002.pgm</image>
  </subtitle>
  <subtitle id="3" start="00:01:13.440" stop="00:01:21.142">
    <image>file10003.pgm</image>

Here is how an example image looked like:

feh after subp2pgm Combine VobSub (sub/idx) Format Subtitles into a Single SubRip (srt) Format File

You can see that the images generated from subtitle2pgm (from the subtitleripper package) didn’t have a grey outline around the characters, where the images generated from subp2pgm do. Actually the subtitle2p2gm utility is a little more flexible and allows you to choose the font and background colors. From the subtitle2pgm read-me page (/usr/share/doc/subtitleripper-0.3/README.subtitle2pgm):

-c <c0,c1,c2,c3>           Override the default grey levels in output image.
                           Default is 255,255,0,255.  Valid values are in the range
                           0<=c<=255 where 0 is black and 255 white.

So in that aspect subtitle2pgm (from subtitleripper) is a little bit better than subp2pgm (from ogmrip)

Convert VobSub Caption Images to Text with transcode

We can use another utility called pgm2txt (from the transcode package), which in turn uses gocr to do the OCR conversion.

$ pgm2txt file1

As the conversion starts, it will show you characters that it doesn’t recognize:

Converting file10767.pgm into text
Converting file10768.pgm into text

# show box + environment
# show box     x=  360    8 d=   3  15 r= 0 0
# list box char:  |(94) l(92) 1(86) I(57)
# show pattern x=  335    6 d=  53  19 t= 1 1
OOO,,,,,,,OOO,,,,,OOO,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
OOO,,,,,,,OOO,,,,,OOOO,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
OOO,,,,,,,,,,,,,,OOOOOO,,###,,,OOO,OOOO,,,,,,,,,,,,,,
,OOO,,,,,,,,,,,,,OOOOOO,,###,,,OOOOOOOO,,,,,,,,,,,,,,
,OOOOO,,,,,,,,,,,,OOOO,,,###,,,OOOOO,,,,,,,,,,,,,,,,,
,,,OOOO,,,,,,,,,,,OOO,,,,###,,,OOOO,,,,,,,,,,,,,,,,,,
,,,,OOOO,,,,,,,,,,OOO,,,,###,,,OOOO,,,,,,,,,,,,,,,,,,
,,,,,OOOOO,,,,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
,,,,,,,OOOOO,,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
,,,,,,,,,OOO,,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,OOO,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
,OO,,,,,,,OOO,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
OOO,,,,,,,OOO,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
OOO,,,,,,,OOO,,,,,OOO,,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
,OOO,,,,,OOO,,,,,,OOOO,,,###,,,OOO,,,,,,,,,,,,,,,,,,,
,OOOOOOOOOOO,,,,,,OOOOO,,###,,,OOO,,,,,,,,,OOO,,,,,,,
,,,OOOOOOO,,,,,,,,,OOOO,,###,,,OOO,,,,,,,,,OOO,,,,,,, -
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
The above pattern was not recognized.
Enter UTF8 char or string for above pattern. Leave empty if unsure.
Press RET at the end (ALT+RET to store into RAM only):

So anything represented by the hash marks (#) it didn’t recognize. You can fill in the characters that you recognize (in the above example that is the i character). After that’s done, you can check how the text looks like :

$ cat file10788.pgm.txt
Still, he's got style.

Combine Converted OCR VobSub Text into a Single SRT with ogmrip

Similar to what we did with srttool (from the subtitleripper package) and .srtx file, we can do with subptools (from the ogmrip package) and .xml file. Here is the command to accomplish that:

$ subptools --convert srt -i file1.xml -o file1.srt -s

We can, of course, check the contents of the resulted srt file:

$ tail file1.srt
787
01:00:24,988 --> 01:00:26,512
What the hell's that?

788
01:00:33,596 --> 01:00:35,461
Still, he's got style.

And of course you can try to fix common OCR issues with the following:

$ sed -f /usr/share/subtitleripper/gocrfilter_en.sed file1.srt > file1_fixed.srt

I did a side by side comparison of the two SRT files, so I ran the following:

$ diff tesseract/file1.srt gocr/file1.srt --side-by-side

and I saw the following:

640                                640
00:48:53,489 --> 00:48:54,510      | 00:48:53,430 --> 00:48:54,454
Hey, you!                          | Hey, you I

641                                641
00:48:54,619 --> 00:48:55,840      | 00:48:54,564 --> 00:48:55,792
I'm talking to you!                | I'm talking to you I

Tesseract did a little bit better. I did another comparison between gocr and tesseract (after using subp2pgm, instead of using pgm2txt I used tesseract). Here is what I saw there:

$ diff --side-by-side ogmrip-tesseract/file1.srt ogmrip-gocr/file1.srt
4                                        4
00:01:22,829 --> 00:01:24,194            00:01:22,829 --> 00:01:24,194
Single file!                             | sin9ie fiiei.

5                                        5
00:01:48,721 --> 00:01:51,417            00:01:48,721 --> 00:01:51,417
Miss, did a strange man                  Miss, did a strange man
just run past here?                      | j-ust run past here?

6                                        6
00:01:51,524 --> 00:01:54,322            00:01:51,524 --> 00:01:54,322
                                         | No
                                         <

It looks like gocr does better with single words (you can see the bottom one, tesseract didn’t recognize the ‘No’, since it’s just a single word). While tesseract does better with long sentences (ie ‘just run’ vs ‘j-ust run’). I saw other similar conversions/mistakes.

Convert VobSub (SUB/IDX) Files into SRT subtitles with vobsub2srt

There is another tool that was recently released that does all the above in one swoop. It wasn’t part of the yum repository so I compiled it manually.

Install VobSub2SRT

First get the prerequisites:

$ yum install tesseract tesseract-devel cmake libtiff-devel

Then get the source:

$ mkdir vb; cd vb
$ git clone https://github.com/ruediger/VobSub2SRT.git
$ cd VobSub2SRT

Configure the package:

./configure -DCMAKE_INSTALL_PREFIX:PATH=/usr/local/vobsub2srt

Prepare the destination folder:

$ sudo mkdir /usr/local/vobsub2srt
$ sudo chown elatov:elatov /usr/local/vobsub2srt

Then build and install the package

$ make
$ make install

Convert VobSub to SRT

Here is the command to do that:

$ vobsub2srt --lang en file1
Selected VOBSUB language: 4 language: en
spudec: Error determining control type 0x38.  Skipping -6 bytes.
SPUasm: packet too short
Wrote Subtitles to 'file1.srt'

The recognition was the best on vobsub2srt (even though both vobsub2srt and subtitle2pgm use tesseract). Here are some small differences that I saw:

$ diff --side-by-side vobsub/file1.srt subrip/file1.srt
36                                   36
00:05:11,711 --> 00:05:14,942        | 00:05:11,769 --> 00:05:15,000
well...                              | we

164                                  164
00:13:58,938 --> 00:14:00,235        | 00:13:58,990 --> 00:14:00,279
Hey, hey-~                           | Hey,he¥

But subtitleripper (subtitle2pgm) ended up getting more captions. If you remember above we had 788 captions, but with vobsub2srt, I ended up with only 786 (still really good):

$ tail file1.srt
785
01:00:24,988 --> 01:00:26,512
What the hell's that?
786
01:00:33,596 --> 01:00:35,461
Still, he's got style.

It looks like it skipped two captions.

Merge 2 SRT into a Single SRT file

This one isn’t that bad. After you converted from vobsub to srt, you should have the following files in the end:

$ ls
file1.avi   file1.srt   file2.avi  file1.srt
file1.idx   file1.srtx  file2.idx  file2.srtx
file1.ps1   file1.sub   file2.ps1  file2.sub

After the conversion, I saw the following for the second .srt file:

$ head file2.srt
1
01:03:03,840 --> 01:03:04,820
Stop there!
2
01:03:06,809 -- 01:03:07,789
Hey, you!

Since the time was correct we can just concatenate the file together like so:

$ cat file1.srt file2.srt > files.srt

You will notice the gap in caption number if you look inside the file:

$ grep ^788 -A 10 files.srt
788
01:00:33,650 --> 01:00:35,510
Still, he's got style
1
01:03:03,840 --> 01:03:04,820
Stop there!
2
01:03:06,809 --> 01:03:07,789
Hey, you!

So we can just renumber our caption IDs, like so:

$ srttool -r -i files.srt -o files-sorted.srt

After that is done, there should be no gap in the IDs:

$ grep ^788 -A 10 files-sorted.srt
788
01:00:33,650 --> 01:00:35,510
Still, he's got style
789
01:03:03,840 --> 01:03:04,820
Stop there!
790
01:03:06,809 --> 01:03:07,789
Hey, you!

Later on, I got a good version of the SRT file and I wanted to compare the times from caption 788 to 789. So I ran the following:

$ grep '^Stop there' files-sorted.srt -B 1
01:03:03,840 --> 01:03:04,820
Stop there!
$ grep '^Stop there' files-good.srt -B 1
01:03:03,779 --> 01:03:04,768
Stop there!

Notice the times are off by a millisecond, this is great. If it’s too far off you can use the subtitle delay functionality from the video player (vlc and mplayer both offer this).

Adjusting the Time in the SRT File

If your second .srt file started from the beginning ie (00:00:00 and not from end of the first file). When I used vobsub2srt on the second file, the resulted file started with the following numbers:

$ head file2.srt
1
00:00:55,335 --> 00:00:56,324
Stop there!

2
00:00:58,304 --> 00:00:59,293
Hey, you!

To adjust the time we can do the following.

Determine Length of Movie

You can use mediainfo for this, like so:

$ mediainfo -f --Inform="General;%Duration%" file1.avi
3728475

that output is in milliseconds, but you can just move the decimal point by 3. Another way is using the ffprobe command, that comes with the ffmpeg package. Here is how that looks like:

$ ffprobe -show_format file1.avi 2>&1 | grep ^duration
duration=3728.478478

Use srttool to Adjust the Time of the Subtitles Inside an SRT file

Now that we know the duration of the first movie, let’s adjust the times in the second SRT file by adding the duration of the first one. Here is the command for that:

$ srttool -d 3728 -i file2.srt -o file2-adjusted.srt

Now we can just concatenate them both and reset the caption ids:

$ cat file1.srt file2-adjusted.srt > files.srt
$ srttool -r -i files.srt -o files-sorted.srt

That’s it :) As a side note, you could play the media files and include the vobsub subtitles like so:

$ mplayer file1.avi -vobsub file1 -vobsubid 4
$ vlc file1.avi --sub-file file1.sub

blog comments powered by Disqus