About

This repo keeps track of which Destiny VODs are currently transcribed for knowyourinfluencer.io

How to contribute

You can contribute to this project by using Whisperer.py and generating a transcript for a video that has not yet been transcribed. All un-transcribed videos have a deidcated issue. The steps are simple:

1. Comment under an issue you wish to claim
2. Complete the transcription process described in the How to transcribe section below.
3. Submit a pull request with the transcript csv file in the transcripts/ directory, addressing the issue you claimed.

How to transcribe

I've made the process as streamlined as I could. All you need is a little programming experience, a GitHub account, and of course the dependencies/hardware requirements needed to run Whisperer.py.

Alternative to dependencies: Docker

If you rather not install the dependencies and are familiar with Docker, you can use the Dockerfile provided instead. Otherwise, go to the requirements section.

I'm not sure on what impact containerizing whisper has on performance, but I would assume it impacts it a little bit. So it might not be the best option, but it's certainly the easiest if you are familiar with Docker.

Assuming you have Docker installed, simply fork this repo, and build & run the container with

# template example
docker-compose build --build-arg video_id=<video_id>
docker-compose up

# real example
# model=base by default, but you can change it yourself depending on your hardware for better performance
# more info at https://github.com/openai/whisper#available-models-and-languages
docker-compose build --build-arg video_id=6zK3i3uK-E0 --model=small
docker-compose up

Requirements

There are 4 requirements for this project.

• Python 3.7 or greater
• ffmpeg
• whisper
• yt-dlp

ffmpeg

You can install ffmpeg using your package manager of choice. (I recommend using a package manager, but you can install from https://ffmpeg.org/download.html instead)

# on Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg

# on Arch Linux
sudo pacman -S ffmpeg

# on MacOS using Homebrew (https://brew.sh/)
brew install ffmpeg

# on Windows using Chocolatey (https://chocolatey.org/)
choco install ffmpeg

# on Windows using Scoop (https://scoop.sh/)
scoop install ffmpeg

whisper

Install whisper

youtube-dl

I recommend installing yt-dlp with pip, which you can do with

python3 -m pip install -U yt-dlp

forking the repo

You'll need to fork the repo so you can later submit a pull request once you've generated a transcript. You can do this by clicking the "Fork" button in the top right corner of the repo page, and cloning it to your machine where you'll be running Whisperer.py.

Running Whisperer.py

Once you've claimed an issue, the video_id you'll be generating a transcribt for is the title of that issue itself. You'll use this when running Whisperer.py.

Whisperer.py will download whatever video you provide, generate the transcript, and output it as <video_id>.csv in the transcripts/ directory. It requires 1 argument, and takes 1 optional argument.

usage: whisperer.py [-h] [--model MODEL] video_id

positional arguments:
  video_id       The Video ID of the YouTube video being transcribed, e.g. "NyiJDDyUV54"

options:
  -h, --help     show this help message and exit
  --model MODEL  The Whisper model being used, e.g. "base" or "large". Depends on your hardware. See https://github.com/openai/whisper#available-models-and-languages to see which models your system can
                 run. Default is "base".

--model is base by default, but depending on your machine's VRAM you can (and probably should) use a larger model.

Make sure you're using the right model for your machine, since using a model that's too large for your machine will cause Whisperer.py to crash.

More info on the model types and hardware requirements can be found here

Example

# using base whisper model
python3 whisperer.py NyiJDDyUV54

# using large whisper model
python3 whisperer.py NyiJDDyUV54 --model large

Submitting a pull request

Once it's run, it will take quite awile to finish depending on the size of the video, so just leave it running in the background for awhile. And depending on the model you chose it could be using a large chunk of your VRAM, which means you'll take a performance hit playing games or other GPU intensive things. So you'll mostly just have to leave it be if that's the case. When it's finished, commit your changes to your repo with a message resolves #xxx, push it to GitHub, and submit a pull request for the corresponding issue you claimed.

Your pull request should be in the following format:

title: resolves <video_id_issue_number>
description: How you want to be listed as a contributor (e.g. what image, name/username), and any notes you have about the transcript.

For example, the following pull request address issue #256

Example

title: resolves #256
description: Username: EvilFossil, image: https://i.kym-cdn.com/entries/icons/original/000/025/526/gnome.jpg

End

If/when your pull request is approved and merged, I'll update the database to include it, and I'll list you as a contributor at https://knowyourinfluencer.io/contributors/.

Thanks for your help. dggL