The live video of the game is interesting and I've been watching it for a long time. There are no days when I don't watch it. Do you know HIKAKIN or Max Murai? It seems that ordinary people know a little about HIKAKIN, which is eaten only on YouTube ... Maybe elementary school students know it.
Anyway, it's interesting, interesting and time-consuming, and it's a substitute for BGM when writing code these days. It used to be anison a little while ago. Maki Daikoku (Is the character different?). Right now, HIKAKIN and Max Murai are the main ones.
If you have time, check out the videos on YouTube. It's not a stemmer.
There are many live video games. There are many newcomers, and finding newcomers is fun. The video is about 5 to 30 minutes long. However, it is a problem that there are too many. There are only 24 hours, a day. You'll want to review old videos as well.
I like watching all the videos, but I also wanted to extract the good points, so I decided to do deep learning. Deep learning ... long. Deep la. Dipla. And.
With Nico Nico, there are comments, and you can know the points of the video's rise by the frequency and content pattern of the comments (wwww, 8888, oh, this is funny and laughing, and applauding). It's easy, isn't it? Download the comment data, and you're done. If you cut out about 10 seconds in the foreground using ffmpeg or something and connect it, you can easily make a digest video and it's fun. I think Ubuntu is good for ffmpeg. Since it is troublesome to install CentOS depending on the version, let's use Ubuntu.
I'm only watching YouTube, but I can't get feedback data from viewers in chronological order like Nico Nico, so what should I do?
Both HIKAKIN and Max Murai have a habit of saying something or a decisive face, and it's a good idea to aim for it and find a point of excitement. I also want to find a screen where Max Murai is squeezed or fluffed.
Deep Learning
Use a Python implementation
$ sudo apt-get install youtube-dl
Introduce a library or tool that downloads all files. After downloading, extract the frame image using ffmpeg and extract the audio considering the bit rate.
For Max Murai's Puzzle & Dragons, a voice like "How about XX!". Does "Hatsudo" work? With HIKAKIN, screaming or strange voices, such voices are prepared for learning data. For the image, capture your favorite face. I feel guilty about learning with HIKAKIN's funny face, but I can't help it because I like it. Max Murai 32-year-old face is also learned separately.
With this, the result that I want to get is that the highlight is extracted by dipping, or that it is in the second and frame of the video.
As I was writing it, it became more and more annoying, so there was no source or command line memo.
If you upload a video of Max Murai's "○○ invocation!" To YouTube, I hope you can remember this article that it worked.
Recommended Posts