16 comments

[ 2.0 ms ] story [ 33.0 ms ] thread
Hi, author of the post here. I thought it'd be nice to post some information about the analysis.

I got some audio features (loudness, energy and valence) for Linkin Park's songs using the Spotify API. In the post, I do an album-wise analysis of these features, and how they vary across and within the albums using the Python Data Analysis Library, Pandas.

I try to articulate the general progression in style through the average change in these features, as well as the change in variation of these features.

If you're a Linkin Park fan, I would love to know how your experience of listening to LP matches with my analysis :)

Excellent work! I'm not a Linkin Park fan although I listened a lot to their albums especially the two first ones. I slowly stopped listening when they released their subsequent albums which correlates with the "energy" loss your data analysis depicts!
Thanks! Same here, listened to the first two albums a lot. Had moved on to other genres by the time they released the other albums. When I tried to go back to LP, there were just too many albums to figure out which one to start with. But after doing this analysis, I feel like I should check out Hunting Party since its the closest to the first two albums.
Perceptual loudness or just some measure or the rms/foote/peakamp? It's proven that these kinds of audio descriptors are frought when trying to attribute them to any human perception.
There is also a "Loudness War" factor in any of popular modern recordings. The dynamic range standards are simply put terrible, despite studies claiming the importance and necessity of keeping the levels good enough.
Fun analysis! It's great that you visualize both mean and std deviation. Having them in two graphs side-by-side makes it a bit hard to read though, as you have to go back and forth.

Using for example a box-plot (https://pandas.pydata.org/pandas-docs/stable/visualization.h...) could show the mean/std in a more comprehensible way by having it in a single plot.

Thanks for the tip! Will keep that in mind going forward.
This is a very cool project! Would you be interested in performing the same analysis of Nine Inch Nail's discography? I am happy to chip in financially for such an analysis if the results would be open.
I didn't know the framework, I seriously hoped it included actual pandas.
Me too. Quite disappointed to be honest.

Was hoping for "We put headphones on 7 pandas and analysed their reaction to Linkin Park's best known songs. 5 out of 7 pandas agreed that the 'Living Things' album wasn't very good."

That is part 2 of the blogpost. I'll be putting headphones on pandas and they'll be outputting the mood of each Linkin Park song.
Interesting to see this here...

One of the things I like about Linkin Park is that their style evolved so much over the years. Even when they got big they were not afraid to completely change up their style almost completely.

I think the author knows this and tries to point it out in the commentary but I don't know if the graphs tell a compelling story.

One thing I'd like to point out from a data perspective is that a bar graph might not be the best tool for this. An album consists of many songs and there are graphs that can convey more information while maintaining readability. Perhaps a scatter plot or candlestick graph? If you use point size for song duration you can even display duration and energy of all the songs on each album in a scatter plot and I think it would be readable still.

I once also analysed a bunch of Linkin Park songs. No matter which algorithm I inputted into the discombobulator, the flux capacitor always overloaded and threw error messages about trash and crap. /me crawls back to reddit
(comment deleted)
PSA if case you're a Linkin Park Fan and in San Francisco... Mike Shinoda (Rapper/Guitarist) is on tour right now and performing in SF tomorrow night (11/6) at the Masonic.