The py-translation project is designed to facilitate language transcription through the Whisper model by OpenAI in a more user-friendly format. Furthermore, the program provides a .srt output file (for timestamped subtitles) and a plain-text file (for reading/writing purposes). The program also includes an inbuilt translation service through Google Translator (potentially more into the future), allowing the end-user to ultimately convert an audio file of any language into translated English text.
Prior to downloading the repository, you require:
- An installation of Python 3.10 or higher
- CUDA Toolkit 11 - Tested with CUDA Toolkit 11.8.0
- cuBLAS for CUDA 11 - Installed through the CUDA Toolkit
- cuDNN 8 for CUDA 11 - Tested with cuDNN v8.9.5 for CUDA 11.x
- An ffmpeg installation with the PATH environment variable assigned (This can be done through Chocolatey with the command
choco install ffmpegor manually through the official FFmpeg site)
After downloading/cloning the repository, run the setup.bat file in the directory. Alternatively, run this in the command-line/terminal to install the requirements for the program:
python -m pip install -r requirements.txt
python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
Run main.py to start the application. The script supports multiple languages for translation however only accepts English as an input for most of its' inputs. You must also have a .wav, .mp3 or .flac file present in the inputs directory prior to running the script (for transcription).
Keys for API-enabled services (currently Microsoft Translator and ChatGPT) must be input into the keys.txt file present in the settings directory following the = sign.
The program can be started through the terminal command:
python3 main.py
By the conclusion of the script, your directory should house one or more of these files:
output.srt(subtitling file, made for SubRip)final_text_output.txt(raw text post-filtering)final_text.txt(like final_text_output, however unique lines only (cleanest file))translated.txt(translated version of final_text, only available after translation is done)
languages.txt contains a list of supported languages for translation of the provided file, including but not limited to English, Spanish, French, and many others.
Contributions are welcome! Please submit pull requests or open issues to propose changes or report bugs.
All rights reserved to appropriate parties, including but not limited to OpenAI, the developers behind stable and fast Whisper, and the parties behind the included libraries.
For support or to report issues, contact pmavro05@gmail.com.