How to find strong excerpts in a transcription with a sliding window in Python
As soon as a transcription has word timestamps, finding short excerpts in a long recording requires no modeling. A fixed-length window, launched at each word and noted according to the number of words that fit in it, finds the passages where the most things are said. The video editor of the osFoundry desktop application offers exactly that with its Suggest clips button, written…
A Python version of osFoundry's "Suggest clips" feature, written in Rust, has been recreated. The feature identifies short clips in a long recording by launching a fixed-length window at each word and rating it based on the number of words it contains. The algorithm was tested on a 5:07 NASA interview with astronaut Christina Koch, which had a transcription of 1007 words.
The results showed that the choice of window length affects the selected clips, with shorter windows rewarding rapid bursts of speech and longer windows averaging out pauses and questions.
Written by urgent.news from Dev.to's report — not a translation of it. Machine-written — may contain errors; check the original before relying on it.
