search

carykh / jumpcutter

Automatically edits vidx. Explanation here: https://www.youtube.com/watch?v=DQ8orIurGxw
MIT License
3.08k stars 541 forks source link

Autodetect and remove verbal filler sounds? #129

Open DMahalko opened 5 years ago

DMahalko commented 5 years ago

Suggestion: Would there be some way to apply this to verbal filler sounds?

ummmmmm uhhhhhhhhhh errrrrrrrrrr Then I guess we should do it this way. Can we, like, talk about this further? We will go ahead and do that step next. The next step of the procedure is to turn around and apply the fluid.

I don't have your genius level of programming skill but I would expect something may be possible via spectrum analysis, and perhaps using AI to look for these specific repeated tone patterns on the audio spectrum of a video.

(If you haven't done this already, it may be worth a youtube project of its own.)

For example, CuriousMarc has a huge collection of interesting videos on antique technology, but English is his second language, and his videos have a horrible long drawn out verbal uhhhhhhhhh as he tries to find the correct things to say..

https://youtu.be/Tvob6u7DbTg?t=12

I would love to experience these videos with those verbal fillers removed.

Lamaun commented 5 years ago

The idea is good, but I think we should keep the project simple. The software you are describing deserves its own project.

If we involve NN here, we will have to collect massive amounts of sound data (which needs to have some labels which parts are filler).

Hardcoding this stuff would be too much effort in my opinion. Also this issue is a duplicate of #125

DonaldTsang commented 4 years ago

Are there any repos that does not use NN though? If we have to, are there any pre-established libraries that does it?

BradKML commented 3 years ago

Back referencing https://github.com/carykh/jumpcutter/issues/122