Capgen is an application that transcribes audio and video using Whisper neural network created by OpenAI. It has a minimal UI that makes it easier for absolutely anyone can use to transcribe or translate all sorts of audio and videos such as podcasts, movies, documentaries, etc. Capgen is also available as a Python command line application here.
For Linux users, you can download the application HERE. After downloading, unzip the archive and run the application executable.
- Source language transcription of audio and video.
- Source language to English translation.
- Tiny, base and small models available.
- Beamsearch and greedy decoding methods are available.
- Support Windows and Mac platforms.
- Provide GPU support.
Capgen is written entirely in C++ and C. It depends on the following libraries:
- FFmpeg: Used to decode media files.
- Libtorch:Performs inference.
- wxWidgets: Provides the graphical user interface.
Capgen is designed so that it can be used on Linux, Windows and Mac but currently, it is tested on the Linux platform only.
Before starting the build process ensure you have the following:
- Cmake: Capgen's build system.
- GNU Make: Builds the application.
- Nasm or Yasm assembler. Required to build FFmpeg.
- Gtk+-3.0: Required to build wxWidgets on Linux.
- Libcurl: Required to allow downloading ability.
- You can install all of these by running:
sudo apt-get install make cmake nasm gtk+-3.0 libcurl4-openssl-dev
Build the application by running the following commands:
git clone --recurse-submodules https://github.com/iangitonga/capgenx.git
cd capgenx/
python3 configure.py
mkdir build
cmake -S . -B build/ -DCMAKE_BUILD_TYPE=Release
cd build/
make
After build process is completed the built application is located in capgenx/bin directory from which you can run and test the application.
If you make changes in the source code, you can rebuild by just re-running make from the build directory.