Open pomkos opened 3 years ago
I lightly read through the code, this seems like a pretty good starting point. Will def have to run a test on some cloud provider's ML cluster to see what the cost is, and how long it takes
The segmenter prints out a list of time segments in seconds of the form (start, end) and saves them as a list of wav files into a folder at
At some point, we will probably have to multithread the algorithm for it to be fast enough for the user.
Their link to the training dataset was broken, here it is: https://catalog.ldc.upenn.edu/docs/LDC97S62/
youtube-dl -f Audio_Only https://www.twitch.tv/videos/982969719 That's what you'll need to do to download the audio streams. Download a few and test to see if it picks out laughter well - https://github.com/jrgillick/laughter-detection It seems like it might have bad results if there is music in the background as well (
@chand1012 and @kjv13 have been working on audio analysis. Tagging them for visibility.
https://github.com/jrgillick/laughter-detection