In my usage yt-dlp gets around the age restrictions. I wouldn’t mind a flag that dumps a list of timestamps instead of having to iterate itself through the video
4 Likes
Yeah, that’s one way. The problem is where to play it since youtube disables embedding age restricted videos. The easiest way would be to find a site that doesn’t care but has a good enough embedded player. Or maybe just hosting it myself. There are some options anyway. And I’ll add a ‘show timestamps’ button in the near future.
1 Like
Not sure if it’ll help but https://www.nsfwyoutube.com/ is able to play 18+ videos:
https://www.nsfwyoutube.com/ watchmore?v={ID}
This is the direct video the website embeds:
www.yewtu.be/embed/ {ID}
2 Likes
I second using an invidious or piped instance (https://yewtu.be/ or https://piped.kavin.rocks/) as an alternative
1 Like
I’ve been going ham using the tool for twitch streams lately and it’s been working really well. Biggest things I’ve been finding kind of funny is when the tool picks up a streamer’s laugh every time or when they play a game with a specific sound affect it picks up every time. Regardless, the tool has been doing good to pick up most burps, even the smaller, more subtle ones.
5 Likes
Yeah it’s kind of funny sometimes. Another one is it’ll pick up every time someone opens a jar or something of that shape.
1 Like
I guess that something, even a random noise, is going to sound more like burping than silence or background static etc. It’ll play back 5% (or whatever that’s set to) of the most likely parts of the video to be burping, even if the most likely parts are just someone talking. Did I understand your question correctly?
It is an RNN. It goes through a wav of audio and as the frequency of audio shifts (as it would on a waveform graph), the program is able to make a prediction of whether the event from a certain segment is a burp, based on a predefined list of samples which are labeled as burps. You are probably noticing that it picks up audio spikes as burps because burps generally move quickly from 1Khz to around 80hz very quickly.
In layman’s terms, the program is doing a little bit of pattern matching as it goes through an audio file to look for burps. The spikes in audio seem to be a trait of the burping sound, which is what you’re noticing
2 Likes
For audio sampling, the industry standard has been some derivative of a recurrent NN, because audio transitions between states, unlike other mediums like picture. Really the best way to make it more accurate is to add more correct samples and reinsert the misidentified events under different labels into the network. Essentially, it’s test correction, but it requires a bit of manual labor.
1 Like
Interesting bc all my YouTube videos are age restricted, but still make it through the scan🤔
I kinda figured that was what was going on. Really, everything that is being sent through could be used to increase the database. Pretty much, the more and longer it is used, should make it more accurate
At the moment the model isn’t improving unless it’s fully retrained, however it could inspire a feature where one could mark segments as accurate or inaccurate predictions can go through training. The only problem I see with this is that rnn training usually isnt incremental and must start from scratch
I looked at that video. I only see the one that is 1 minute and 50 seconds from 1 year ago. Nothing in that video, it’s like a trailer and there are no burps at all.
There was a burp in there, it was the one I sent on the private chat
@Q what is the input when your tool scans twitch streams? Would it be able to scan sub-only vods too with an m3u8 link?
Possibly if the video could be accessed somehow. @TC, do you know if that download trick you found still works?
my chat_reader tool gets m3u8’s for channels, you could possibly use sed to filter its output to a single vod. Here’s the way to do it for individual vods with the raw html (Haven’t done this in a long time, but it most likely still works): Downloading twitch vods from channels that don't allow clipping - #2 by TC
You can curl the page and it should still have the same information found in the html
Thanks, @TC. I checked quickly and it seems to still work. @MrBaddie, one way to do that for the time being could be to download the video as described in TC’s link and then reupload to an unlisted youtube video and then submit that link. That might work.
1 Like
Alright, after a few weeks using the tool more, one of my biggest feature requests is to make the tool ignore laughing. Absolutely no idea if it’s even possible, but if the tool could ignore laughing, that would be incredibly beneficial. Laughing seems to be one of the most common things the tool picks up that isn’t burps. It also drives me crazy when listening to clips back to back and the twitch streamer will laugh in the exact same way every time. Almost like a recording stuck on repeat. 
2 Likes
I guess laughing in those videos sounds most like burping compared to anything else there and so it will show you that. It also tends to happen with coughing for example and similar kinds of sounds. You might be able to decrease the amount of laughing you hear by setting the percent to something lower.