VO-Transcriber

A transcriber for VOs (Vorlesungen / Lectures) for Students of TU Wien and Uni Wien

This is a transcriber that uses Whisper (by OpenAI) to transcribe recorded lectures from the TU Wien and Uni Wien. It is a docker container that can be run on any machine that supports docker and has it installed.

I noticed that some of the lectures at TU Wien and Uni Wien have poor or even no scripts available. This is a problem for students that want to learn from the lectures but don't have the time to watch them. This project aims to help solve this problem by providing transcripts for the lectures. Of course, this is not a perfect solution, but used in combination with some new AI-Tools like ChatGPT or the Bing Chat Bot, it can help students to easier consume lectures and use this scripts as a lookup place.

Installation

Note: You need to have docker installed on your machine. If you don't have docker installed, you can find the installation instructions here.

I do not recommend to install this project on your own machine and rather put this project onto a server. The reason for this is that the transriber uses a lot of resources and even only one transcription could take many hours to complete. If you want to use your own machine, you can but you are warned.

1. First of all you need to clone this repository to your local machine:

   git clone git@github.com:bananensplit/VO-Transcriber.git
   cd VO-Transcriber

2. Before you build the image please check in the Dockerfile if the correct lines are commented out. Depending on your architecture you need to uncomment the correct lines.

   # This is for ARM64
   RUN wget https://github.com/wkhtmltopdf/packaging/releases/download/0.12.6-1/wkhtmltox_0.12.6-1.buster_arm64.deb && \
   	apt install -y ./wkhtmltox_0.12.6-1.buster_arm64.deb && \
   	apt install -y openssl build-essential libssl-dev libxrender-dev git-core libx11-dev libxext-dev libfontconfig1-dev libfreetype6-dev fontconfig
   
   # This is for AMD64
   # RUN wget https://github.com/wkhtmltopdf/packaging/releases/download/0.12.6-1/wkhtmltox_0.12.6-1.buster_amd64.deb && \
   #     apt install -y ./wkhtmltox_0.12.6-1.buster_amd64.deb && \
   #     apt install -y openssl build-essential libssl-dev libxrender-dev git-core libx11-dev libxext-dev libfontconfig1-dev libfreetype6-dev fontconfig

   To check which architecture you have, you can run the following command

   uname -m  # on Linux
   $Env:PROCESSOR_ARCHITECTURE # on Windows

3. Then you can build the image:

   docker build -t vo-transcriber:0.1.0 .

Usage

The following section(s) contain(s) information on how to run the container and how to set the correct parameters.

Parameters

• --help - If this parameter is set, the help will be printed.

• --uni - The university you want to transcribe the VOs from. Possible values are tu and uw.

• --vos - The name of the VO you want to transcribe. This parameter can be used multiple times. If you don't provide this parameter, nothing will be transcribed.

• -p - The path to the file containing the VO-data. Can be used with the --uni parameter set to tu or uw

• -k - The link to the VO-data. Can be used with the --uni parameter set to uw.

  usually the link to the data looks something like this: https://ustream.univie.ac.at/search/episode.json?limit=200&offset=0&sid=XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX

• -m - The name of the Whisper model you want to use for transcribing. Possible values can be looked up on the Whsiper Github.

• -v - If this parameter is set, the verbose parameter will be passed to whisper and you will be able to see realtime translations.

• -l - The language of the VO. Possible languages are listed here. Default is de.

• --txt - If this parameter is set, the transcription will be saved as a txt file.

• --vtt - If this parameter is set, the transcription will be saved as a vtt file (format for subtitles).

• --srt - If this parameter is set, the transcription will be saved as a srt file (format for subtitles).

• --pdf - If this parameter is set, the transcription will be saved as a pdf file.

• -o - The output folder. Must be output (Docker and stuff).

Startup - VO-Data from link

Note: This only works for the Uni Wien. The TU-Wien (unlike the Uni Wien) has authentication which complicates things a bit. You can't provide a link to the data because this project doesn't support authentication (yet).

Before you start make sure the link you provide is correct and public (no login is required to retrieve the data).

1. To look up which vos are available, run the following command:

   docker run --rm --name vo-transcribe \
   	vo-transcriber:0.1.0 \
   	--uni "uw" \
   	-k "https://ustream.univie.ac.at/search/episode.json?limit=200&offset=0&sid=XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX"

2. When you choose which VOs you want to transcribe, run the following command:

   mkdir output
   
   docker run -d --name vo-transcribe \
   	-v $(pwd)/output:/usr/src/app/output \
   	vo-transcriber:0.1.0 \
   	--uni "uw" \
   	-k "https://ustream.univie.ac.at/search/episode.json?limit=200&offset=0&sid=XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX"
   	--vos "<name of the VO you want to transcribe>" \
   	--vos "<second VO you want to transcribe>" \
   	--vos "<... VO you want to transcribe>" \
   	--vos "<last VO you want to transcribe>" \
   	-o "output" \
   	-m "medium" \
   	-l "de" \
   	--txt \
   	--vtt \
   	--srt \
   	--pdf

You can of course adjust the parameters to your needs. This command would

• transcribe 4 VOs
• use the medium model (takes about 3 hours for 1,5 hours of VO)
• use the de language
• save the transcription as txt, vtt, srt and pdf files

Startup - VO-Data from file

Note: In the following instructions the --uni parameter is set to tu (TU Wien) but the same steps would also work for Uni Wien (uw).

1. Create the mount folder and the data file which contains the VO-data.

   mkdir ouput
   echo "data" > output/data.json

2. To look up which vos are available, run the following command:

   docker run --rm --name vo-transcribe \
   	-v $(pwd)/output:/usr/src/app/output \
   	vo-transcriber:0.1.0 \
   	--uni "tu" \
   	-p "output/data.json"

3. When you choose which VOs you want to transcribe, run the following command:

   docker run -d --name vo-transcribe \
   	-v $(pwd)/output:/usr/src/app/output \
   	vo-transcriber:0.1.0 \
   	--uni "tu" \
   	-p "output/data.json" \
   	--vos "<name of the VO you want to transcribe>" \
   	--vos "<second VO you want to transcribe>" \
   	--vos "<... VO you want to transcribe>" \
   	--vos "<last VO you want to transcribe>" \
   	-o "output" \
   	-m "medium" \
   	-l "de" \
   	--txt \
   	--vtt \
   	--srt \
   	--pdf

You can of course adjust the parameters to your needs. This command would

• transcribe 4 VOs
• use the medium model (takes about 3 hours for 1,5 hours of VO)
• use the de language
• save the transcription as txt, vtt, srt and pdf files

Setup for development

This are instructions for people who might want to contribute to this project on how to get this project running on their local machine:

Note: This project uses Python 3.10 and i recommend to use virtualenv to manage all the dependencies.

Setup

# Clone this repository
git clone git@github.com:bananensplit/VO-Transcriber.git
cd VO-Transcriber

# Create a virtual environment
python -m virtualenv venv
source venv/bin/activate

# Install the dependencies
pip install -r requirements.txt

Run the project

This is as far as the setup goes. You can now run the project with the following command:

python main.py
# The main.py takes the same parameters as the docker image
# When no parameters are provided, the program will return its help message

Build the docker image

To build the docker image, run the following command:

docker build -t vo-transcriber:0.1.0 .

Find a bug? Have an idea?

If you find a bug in the source code or a mistake in the documentation, you can help me by submitting an issue in the Issuetracker. Even better you can submit a Pull Request with a fix.

Furthermore if you have an idea for a new feature, feel free to submit an issue with a proposal for your new feature. Please add as much detail as possible to the issue description. This will help me to understand your idea and to discuss it with you.

Thanks for making this project better!

Contact

Jeremiasz Zrolka - jeremiasz.zrolka@gmail.com

• Twitter: @jeremiasz_z
• Instagram: @jeremiasz_z
• LinkedIn: jeremiasz-zrolka