Introduction

Extract Sub from Video for Streamlit : Extract hardcoded subtitles from videos and Youtube.

Using a Streamlit interface, Youtube video download + video cropping and subtitle extraction.

Program works with machine learning, multilingual OCR toolkits, ffmpeg, a flask server, docker...

##what technology and program used

• PaddleOCR (PaddleOCR aims to create multilingual, awesome, leading, and practical OCR tools that help users train better models and apply them into practice.)
  • Paddle works on docker because it requires a specific version of Linux. But a flask API is developed to make it work. You just have to launch the docker command. See the explanations about the installation.
• Videocr Extract hardcoded (burned-in) subtitles from videos using the PaddleOCR OCR engine with Python
  • videocr is integrated in the flask api which is launched with the docker.
• Streamlit a graphical interface allows to manage all the process of file upload, Youtube video download as well as cropping and subtitle extraction.

##how to install it

Note: you must have docker and python3.7 minimum installed on your machine

sudo docker build -t your_name .
sudo docker run -p 5000:5000 --name your_name_container your_name

From there, you should have the api that handles OCR working well

cd app
pip install -r requirements.txt
streamlit run app_streamlit.py

Note: modify pip by pip3 for example depending on what you have

Streamlit server starts, you can start using the application and play with the settings.

API

1. filename

   File name uploaded to the docker server, either a cropped video or an original video

2. lang

   The language of the subtitles. (French, Chinese, English, Latin)

3. gpu

   if 1 the OCR process uses the GPU (it is more powerful and faster than the CPU.

4. sim

   Similarity threshold for subtitle lines.

5. conf

   Confidence threshold for word predictions. Words with lower confidence than this value will be discarded.

6. start_time and end_time

   Extract subtitles from only a clip of the video. The subtitle timestamps are still calculated according to the full video length.