/ffmpeg-slide-gen

Combine slides and audio to produce video using ffmpeg

Primary LanguagePythonApache License 2.0Apache-2.0

ffmpeg-slide-gen

Combine slides and audio to produce video using ffmpeg

What is this repo about?

This repo provides a Python script to automate the execution of ffmpeg in order to produce a presentation video. It takes the following as input:

  • A sequence of slide images
  • The audio of the presentation
  • Timestamp for each of the slides

The script then computes the durations for each slide to produce a concat file and runs it through ffmpeg along with audio to generate a presentation video.

This makes it easy to record the audio of a presentation (the easiest part) and later combine it with relevant presentation with ease.

Requirements

Tested with Python 3.9.1 but may work with other Python 3 versions. Requires a few Python libraries that can be installed (virtualenv is preferred) using the command:

pip install -r requirements.txt

Also depends on ffmpeg and mediainfo commands so make sure to have them in the shell PATH. Tested with the following versions:

  • ffmpeg 5.1.2 (can be installed using Homebrew by running brew install ffmpeg)
  • mediainfo v22.12 (can be installed using Homebrew by running brew install mediainfo)

NOTE: All the testing was done on Mac OS Ventura 13.1

Usage

Run the ffmpeg-slide-gen.py script with --help to see the usage. All options are required (except --help itself).

Usage: ffmpeg-slide-gen.py [OPTIONS]

  Generate FFMpeg demux concat file from specified slides and timestamps. See:
  https://superuser.com/a/619843/26006

Options:
  -c, --slide-count TEXT          Number of slides.
  -i, --slides FILE...            List of image filenames as an ordered
                                  sequence.
  -t, --text-timestamp, --txt <TEXT TEXT>...
                                  Pair of: <timestamp slide-text>. Timestamp
                                  can be in seconds or [hh:]mm:ss. Do not mix
                                  with --slide-timestamp/--vlc-playlist-file
                                  options.
  -s, --slide-timestamp, --ts <INTEGER TEXT>...
                                  Pair of: <slide-number timestamp>. Timestamp
                                  can be in seconds or [hh:]mm:ss. Not needed
                                  if --vlc-playlist-file is specified.
  -b, --vlc-playlist-file, --m3u FILE
                                  Specify the path to the m3u file to extract
                                  bookmarks as slide timestamps. Not needed if
                                  --slide-timestamps is specified.
  -a, --audio-file FILE           Specify the path to the audio file.
                                  [required]
  -o, --video-out TEXT            Specify the path for the output video file.
                                  [required]
  -vs, --video-scale TEXT         Specify an alternative scale for output
                                  video  [default: 1280:720]
  -n, --dry-run                   Enable dry-run mode, generates the concat
                                  file and prints the ffmpeg command without
                                  actually running it.
  --help                          Show this message and exit.

Usage notes

  • --slides option takes multiple image file names as arguments, but the count of the filenames must be specified using --slide-count option.
  • --slide-timestamps and --vlc-playst-file are mutually exclusive options as they are meant to specify the slide timestamps. Specify only one of these options.
  • --slide-timestamps takes a pair of arguments, first being the slide number and the second being the timestamp of the slide. The timestamp can be given as seconds (integer), seconds.millis (float) or min:sec (string).
  • --dry-run option runs all the logic, but instead of running the ffmpeg command, it is simply echoed to the console. In dry-run mode, the temp concat file will not be deleted, so the echoed ffmpeg command can be executed as is.

Overall Approach

Approach 1

  • Export presentation as images
    • A PPT file can be opened in Keynote and all slides can be exported to a folder, say /tmp/proj
    • A PDF can be exported to images using ghostscript, e.g., gs -sDEVICE=jpeg -sOutputFile=page-%03d.jpeg -r100x100 -dNOPAUSE -dBATCH -f file.pdf
  • Generate an interim video by running ffmpeg-slide-gen.py with just the first slide
  • Open the presentation and the intermim video
  • While playing the video, add a bookmark at each location when a slide starts and rename it with the slide number
  • At the end save it as a playlist which should create a .m3u file
  • Rerun ffmpeg-slide-gen.py with the m3u file as input to produce the final video.

Sample commands

$ python ffmpeg-slide-gen.py -a /tmp/proj/audio.mp4 -o /tmp/proj/video-tmp.mp4 -c $(echo /tmp/proj/*.jpeg | wc -w) -i /tmp/proj/*.jpeg -s 1 0
$ cat /tmp/video.m3u
#EXTM3U
#EXTINF:2438,video-tmp.mp4
#EXTVLCOPT:bookmarks={name=1,time=0},{name=2,time=5},{name=3,time=35},{name=4,time=40},{name=5,time=105},{name=6,time=476},{name=7,time=719},{name=8,time=832},{name=9,time=892},{name=10,time=904},{name=11,time=965},{name=12,time=979},{name=13,time=1068},{name=14,time=1167},{name=15,time=1178},{name=14,time=1261},{name=16,time=1310},{name=15,time=1406},{name=17,time=2139}
file:///private/tmp/video-tmp.mp4
$ python ffmpeg-slide-gen.py -a /tmp/proj/audio.mp4 -o /tmp/proj/video.mp4 -c $(echo /tmp/proj/*.jpeg | wc -w) -i /tmp/proj/*.jpeg -b /tmp/video.m3u

Approach 2

  • Export presentation as images
    • A PPT file can be opened in Keynote and all slides can be exported to a folder, say /tmp/proj
    • A PDF can be exported to images using ghostscript, e.g., gs -sDEVICE=jpeg -sOutputFile=page-%03d.jpg -r1280x720 -f file.pdf
  • Open the presentation and play the audio to identify the timestamps
  • Whenever a change in slide is observed, take a note of the slide number and timestamp (in seconds)
  • Use the above timestamps to construct the command-line to produce final video

Sample command

$ python ffmpeg-slide-gen.py -a /tmp/proj/audio.mp4 -o /tmp/proj/video.mp4 -c $(echo /tmp/proj/*.jpeg | wc -w) -i /tmp/proj/*.jpeg \
  -s 1 0 \
  -s 2 5 \
  -s 3 35 \
  -s 4 40 \
  -s 5 105 \
  -s 6 476 \
  -s 7 719 \
  -s 8 832 \
  -s 9 892 \
  -s 10 904 \
  -s 11 965 \
  -s 12 979 \
  -s 13 1068 \
  -s 14 1167 \
  -s 15 1178 \
  -s 14 1261 \
  -s 16 1310 \
  -s 15 1406 \
  -s 17 2139

As can be seen above, slide numbers can be repeated and in any order, in case the presenter doesn't go through them sequentially or goes back and forth.

Addendum

This repo also comes with the youtube-transcript.py script that can be used to retrieve autogenerated transcript from a YouTube video. It is most useful to do the following:

  • Uload the interim video file generated in Approach 1 to your YouTube account as an unlisted video.
  • Wait for a few hours for YouTube to generate the transcript for the audio from the presentation.
  • Run the tool to dump the transcript to an HTML file, e.g., python youtube-transcript.py -d 0 -s abcdefghijk
  • Open the HTML file in a browser and use that as a reference to locate the timestamps for slides, e.g., you can look for certain keywords like "slide" (when the presenter speaks of slide numbers) or slide titles and unique keywords in the slide content.

Usage

Usage: youtube-transcript.py [OPTIONS] VIDEO_ID

Options:
  -d, --interval-duration, --id INTEGER
                                  Interval duration for generating timestamps,
                                  use 0 to disable and produce raw output
  -s, --timestamp-in-seconds, --tis
                                  Specify to see duration in seconds instead
                                  of min:sec
  --help                          Show this message and exit.

NOTE: the above sample commands rely on the shell wildcard expansion to order them sequentially, which works as expected for the slides exported from Keynote on a Mac computer.