Information Density based video speed selection


Watch videos and change their speed

YAML ソース 発想

I sometimes listen to Youtube videos as I do other things that totally are not playing video games. I have a permanent frustration with Youtube videos. Some parts that have interesting, informative, or otherwise important content go by too quickly, and sometimes the information is repetitive, introductory, basic, or otherwise non-salient and go by too slowly. I don't want to spend 20 minutes listening to the words "like and subscribe" and leading into a video with stuff I've heard a million times before. But I don't want to stop what I'm doing, find the tab playing the video, and click around until the actual content starts. And I don't want to do the same to go back to the beginning of the important or interesting bit of a video.One tool Youtube has is the speed selection button. I sometimes set the speed high until the introductory part is over, listen to an information-sparse video on high speed, or listen to a particularly information-dense video on low speed. I want a program that can do some of this for me. Some parts of salience can be automatically detected. Erring toward normal speed a program can speed through introductory content the same way I do: Set the speed to 200% until words are heard similar to "without further ado" or "and this brings us to the important part". It's also possible without strong AI to observe word length and commonness. As information density grows so does, usually, word length. Furthermore the speaker's talking speed can be evened out. If he's excited and talking fast the words per minute can be smoothed out using a smoothing algorithm to set the whole video at a particular number of words per minute.Such a program is proposed either for individual usage or as part of a video streaming service."Drunk mode" sets the information/time density to 50%. "information search mode" runs the video through a text to speech converter, outputs the text, and speeds through everything not surrounding unique or explicitly searched-for words.

Credits: Voice of HalfBakery.

I think this idea somehow relates to the Resampling Summarizer for Social Media.

(通知しない) (不必要) ログインしてください。