calcoloergosum/video-watching-acceleration

Video Watching Acceleration

Better experience for watching videos where people talk.

This script was intended to watch small scale math lectures/seminars, usually hours long.

Sample

Before: 28 seconds (Click to play):

After: 20 seconds (Click to play):

Motivation

I watch videos fast-forward, usually x1.25. Faster than that I cannot fully comprehend the utterances anymore. But when presented together with subtitles, I can comfortably follow the argument of the speaker with playback speed of x1.5 or more (depends on the speaker of course). Also, I usually skip the pauses by pressing right-arrow button on the keyboard with VLC. But it is easier to pause when I can't follow the video, than skipping when I can fully follow the video.

Personal result is:

• Speed up by subtitles: about 25%
• Speed up by long pause skip: about 60%
• Overall speed-up: about 200%

Methodology

1. Prepare subtitle

   Whisper by OpenAI(MIT license) was used. To better adjust timestamps, stable-ts was used.

2. Clip long pauses

   Silero-VAD(MIT License) was used.

3. When watching, fast-forward according to your preference

   Any media player with playback speed control will do; e.g. VLC.

How to use

After installing requirements by pip install -r requirements.txt,

Start with a directory that contains videos.

data_src/
    - lecture01.mp4
    - lecture02.mp4
    - ...

Run python -m video_watching_acceleration.transcribe mp4 data_src data_src --whisper-model small

data_src/
    - lecture01.mp4
    - lecture01.srt
    - lecture02.mp4
    - ...

Run python -m video_watching_acceleration.vad mp4 data_src data_src

data_src/
    - lecture01.mp4
    - lecture01.srt
    - lecture01_vad.csv
    - lecture02.mp4
    - ...

Finally, run python -m video_watching_acceleration.edit mp4 data_src data_dst

data_src/
    - lecture01.mp4
    - lecture01.srt
    - lecture01_vad.csv
    - lecture02.mp4
    - ...
data_dst/
    - lecture01.mp4
    - lecture01.srt
    - lecture02.mp4
    ...

Enjoy!