Part 3

https://devrant.com/rants/9881158/...

I dropped subtitles and started extracting audio from movie, after that I use whisper to convert speech to text.

I parse srt from whisper, adjust timestamps to get >= arbitrary amount of voice seconds. I put text to vector database with timestamps and movie file name.

I query database by ex. “I don’t know” and extract first n results, after that I walk trough movies and extract parts with found text.

I normalize and merge parts into one movie.

Results are satisfying so now I decided to try to find a common dialogue that I can watch by combining multiple persons speaking from multiple movies.

Might also try to extract person from one movie and put it to other movie.