Prerequisite

It’s possible during the process of installing different things you have to close and reopon your powershell or command prompt window. In case you run into a lot of trouble of commands not working just restart your PC to be safe.

• Install GIT
  • During installation mare sure to select to add it to PATH as well
• Install Python 3.10
  • During installation make sure to select to add it to PATH as well
• Install FFmpeg
  • From the website select this one ffmpeg-master-latest-win64-gpl.zip (or select the one applicable to your OS)
  • Unzip the folder and copy the folder to your C:\
• Install NVIDIA CUDA Toolkit
  • Check version with nvcc —version in a command prompt (if you receive a version output it means it is correctly installed)
• Add system environment variables

  • Open start menu

  • Search for “Edit the system environment variables”

  • Open the “Environment Variables” and make sure these are there or add them to the User variables (double click to open PATH or click Edit) NOT system variables

    

    

    • C:\Users\<yourusername>\AppData\Local\Programs\Python\Python310\Scripts\

    • C:\Users\<yourusername>\AppData\Local\Programs\Python\Python310\

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.5\bin\

    • C:\FFmpeg\bin\

Libaries & Repo’s

Open a command prompt for the following commands

• WhisperX >> used to transcribe and align

  pip install git+[<https://github.com/m-bain/whisperx.git>](<https://github.com/m-bain/whisperx.git>)
  

• Ffsubsync >> Syncs the subfile either with video or other srt file

  pip install ffsubsync
  

• Pytube >> downloads youtube video and audio

  pip install pytube
  

• Torch, torchvision, torchaudio (Check latest supported cuda version)

  pip install torch torchvision torchaudio --index-url <https://download.pytorch.org/whl/cu121>
  

  • Check if GPU processing is available

    • Create a new file called torch_version.py
    • Add the following piece of code

    import torch
    
    print(torch.__version__)
    print(torch.cuda.is_available())
    

    • In the folder where the file is created SHIFT + Right Click and open PowerShell or ALT + D and type CMD
    • Start the script by using

    python .\\torch_version.py
    

    • The output should like like

    2.3.1+cu121
    True
    

    • You can remove the file afterwards

How to use

There are several ways to make use of whisperX to create subtitle files for whatever you are watching, or even listening to. I will list different possibilities down below which you can choose to follow and/or combine.

There are different options to use for the command, in case you are interested in that please look them up online.

Basic usage

Put the file you want to let whisperX transcribe into a folder of your preference. From within that folder open a PowerShell window (SHIFT + Right Click in the folder or ALT + D and type CMD) and paste the following command.

whisperx "filename" --model large-v2 --language Korean --batch_size 1 --compute_type float32 --device cuda

What this does is create a .srt (subtitle file) and some other files based on the specified file. It is possible the subtitles dont match the timing, and  there are some possibilities to fix that which I will describe in other sections below.