Build a dataset from YouTube

I work to build dataset from YouTube as audio-visual dataset for lip reading propose but i faced some issues in the my code when i want trying add shot boundray detection before face detection and tracing how i add whole those things together and after that using active speaker detection