Detect and extract hardcoded video subtitles into subtitle srt file
INSTALLATION Make sure you have CUDA and cuDNN installed on your system Ideally CUDA 11.7 and cuDNN v8.4.1 If you have CUDA 11.7 already, download cuDNN v8.4.1 from : https://drive.google.com/file/d/1L3PwFszUul6l4xyeMAEsttQtcLFOlTgh/view?usp=sharing -Extract the cuDNN folder -Copy the bin,include,lib folder to "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.7"
-
Install PaddleOCR (PaddleOCR is a state-of-the art library for text detection and recognition)
1.1 Download PaddlePaddle whl file from : Windows 11,CUDA 11.7 https://drive.google.com/file/d/1PAuRX-tm6hBbY0cB0w5jegFg42NVqWNa/view?usp=sharing Windows 10,CUDA 11.7 https://drive.google.com/file/d/1DiNpU-AIE8rwa5VldAEds-ySiRQ1bc8K/view?usp=sharing
For other versions visit: https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/windows-pip.html
1.3 From the download folder run:
pip install 'downloaded .whl file'1.4 Install PyMuPDF manually (need v1.20.2. pip install casues problem sometimes)
Download PyMuPDF: Windows 10: https://drive.google.com/file/d/1BsTqOrQJGZbDb_arNy0IwJjrrS6pjN34/view?usp=sharing Windows 11: https://drive.google.com/file/d/1AQuhBaS23tW8y0iJwL1YltIhcRfzI4fy/view?usp=sharing
Install:
pip install 'downloaded .whl file'Install paddleocr:
pip install paddleocr>=2.7 -
Install other dependencies
Run from the main code directory
pip install -r requirements.txt
RUNNING
Open terminal and run this command:
python main.py --video_path "VIDPATH" --save_dir "SAVEDIR" --language LANG --start_time "00:00:00" --end_time "00:10:00"
replace VIDPATH with path to video file
replace SAVEDIR with path where you want the srt file to be saved. Just give directory path. The saved file name will be 'subtitles.srt'
replace LANG with the language of the subtitles in the video. Use abbreviations. E.g. For english use "en" . For other languages see the supported_languages.png file
start_time and end_time are the time for which the subtitle are made. Format is HH:MM:SS
Additionally, use --use_gpu=True flag to do processing using GPU
Note: During first run with a new language, the model files for that language will be downloaded automatically.