How to fasten the text extraction of a video

[onkar-hingne-nice]

i want to extract all the text in the video. I am first unpacking all frames and then finding dissimilar frames by doing a hash comparison with a very strict threshold. Then I run easy OCR on all the dissimilar frames to extract the text from those frames. I am using GPU on AWS g6.2xlarge to do this easyOCR execution. it takes upto 30 secs to complete this activity. How can i reduce the the time taken for this extraction?

[onkar-hingne-nice]

I was cheking other options lets say I am not able to reduce time taken for one video but if I can run multiple videos in parallel on the same GPU even then its agood for me since i am able to increase my throughput. But when i run parallel processing of the easyocr in process pool I am getting bad result its taking more time to complete.