The Quest to automatize the video translation of lectures

Here we define a video lecture, as a video with someone talking about a subject and using a powerpoint presentation. This type of video is used in every chapter of any courses in PlanB Network, so there's a need to find a way to make the whole process as automated as possible.

Video Translation workflows

• inputs: video vo + ppt vo of n slides
• process 1: extract n audios from videos vo
• process 2: translate audio
• process 3: translate ppt
• process 4: export ppt into images
• process 5: generate video from n images and n audios, both translated
• output: video translated

Backlog

• ppt vo -> ppt translated
• mp3 vo -> txt vo -> txt translated -> mp3 translaled
  • use of audio_translator.py
• ppt -> png
  • use of ppt_2_png.sh
• png + mp3 -> mp4
  • use f image_audio_2_video.py

Requirements

• only tested on linus distribution
• for bash scripts you need this package installed
  • LibreOffice - 24.2.5.2
  • ImageMagick - 6.9.12.98
• for python scripts, just create a dedicated environnment and use the command pip3 install -r requirements.txt to install the required librairies

---

Audio Translator

Initial Goal

Develop a Python program to automatically transcribe all segmented audio files using a text-to-speech API, simply said an audio-2-audio Bot with translation in between.

Initial Strategic Plan

• write a single file called playground.py for quick MVP, though constant will be placed in config.py
• encapsulate atomic process using simple functions
• will use LLM-Translator and LLM-Scribe if needed
• use folder projects to centralize all file produced
  • /projects/project-name/audios/: initial audio to translate
  • /projects/project-name/transcripts/: corresponding transcripts
  • /projects/project-name/translations/: corresponding translations
  • /projects/project-name/outputs/: audios produced with 11labs api