Analyzing the chords of 1300 popular songs for patterns
hooktheory.com
hooktheory.com
https://briancort.com/songviz/
I also made a visualization to explore chord harmonies:
It was driving me mad -- a few times I caught myself during work hours trying to reproduce it from memory in MuseScore. It was finally played fully in the last episode, and the aha moment was that what appeared to be a minor chord before The Chord was a major seventh and the progression resolved neatly. What it seems to have been done is to play multiple variants that played component notes differently at different velocities (the strength with which you hit a piano key). It's like it moved by aporias, each time leading you into a different impasse. The show's plot is sort of like that too.
This is to say -- there's more to music than chord progressions, and there's more to chord progressions than chords :)
Still, don't want to be too negative, it's very interesting that he used data from guitar chord websites, that's a clever data source.
>Guitar tab websites have tons of information about the chord progressions that songs use, but the quality is not very high.
>So, over the past 2 years we’ve been slowly and painstakingly building up a database of songs taken mainly from the billboard 100 and analyzing them 1 at a time.
Seems like a basic statistical analysis of songs. Anyone who has composed or even played music will find no surprises here (unless you solely loved on atonal 20th century stuff).
However, the author clearly has knowledge of music theory, so I fully fail to appreciate the purpose of the article.
Emin vs Emaj - my gut would say Emin is going to fit in with keyboard/piano stuff easier, and there's more of that type music overall. When I think Emaj, I think guitar stuff (EAD stuff), and there's been less of that overall compared with keyboard-based music.
It’s useful for dreamy and somewhat static chord progressions though.
D/E/A - again, ime - were always "guitarish" songs. GCF always seemed to be more "pianoish". Just my own observation/experience - it may not be true of everything, but seems to be true about much of the stuff I've listened to (and tried to play).
A major (A/D/E) is probably the easiest key to play. G major is arguably the most useful, because you have access to all the important chords in open position: G, Am, C, D, Em.
Also, in this way we can say "C major" or "C minor" to refer to the key of the song. We use "natural" when its important to distinguish from a note that has a sharp or flat - e.g. C# versus C natural.
In Italian, how would you refer to the key of Db major - "Db natural"? How would you refer to the key of Db minor?
By that measure it's only natural that Em (iii) is more common than E (which is non-diatonic to C).
(Also, songs are often written in one key, then transposed in the studio.)
I'm not sure why F would be more popular than D and A. One hypothesis would be that it has fewer accidentals; it's not a fun key for guitars, but many contemporary pop songs aren't guitar-driven anyway.
EDIT: Thinking about it, E minor is fairly common in phrygian scale (it's the tonic chord), however, phrygian, unlike dorian or myxolidian, which are used at least sometimes, is super rare -- its dominant chord, B dim, is dissonant, which kind of sucks.
Functional harmony kind of goes out the window once you stray from the major/minor modes anyway. You have to lean heavily on the root to maintain the tonality, or else the perceived key center tends to shift.
I think it's safe to say the effect of metal songs using phrygian is negligible.
It would have been nice if the author also had analyzed chord progressions instead of single chords.
Who knows how many Pachelbel's progressions he would have found! :)
Last heading in this first section:
“If a song happens to use a particular chord, what chord is most likely to come next?”
I meant something like I-V-vi-iii-IV-I-IV-V
God forbid your song deviates from the most 1300 most popular songs!
Great work though! I really like the clear and informative data displays.
There is a ton more to music than chord progression. Figuring out which parts to change, and how, and which parts to leave alone is key.
If you're writing pop, you want it to be a hit, and you want to use a non-standard chord progression? You better know your harmony theory inside out. Otherwise stick to the usual chord progressions.
I guess it's like saying "if you want to build a native iOS app, you should probably think very hard if you want to use something other than Objective-C or Swift to build it"
If that’s not “pop” enough, I’m pretty sure it’s the same structure McCartney used in “Yesterday”.
Which, if you want to see how to break every damn rule and still make hits, ladies and gentlemen: The Beatles!
Good call on the guitar tab sites being wrong almost all the time!
> It works fine for songs that modulate.
By definition, modulation is the act of not remaining diatonic. Modulation = changing keys. Diatonic = staying within key.
So, no, we don't need to assume anything about diatonicity in order to successfully do harmonic analysis using Roman numerals.
https://www.hooktheory.com/trends
It’s fun to play with and has helped me find songs to hear how the same progressions can be different across different songs. Personal favorite progression is I-IV-VI-V...which is shared by “Whatya Want from Me” by Adam Lambert and “Concerning Hobbits”.
I normalized it such that "key" was irrelalent because transposed keys are generally considered the same such that I was more concerned about the relative relationship between chords per song.
After the analysis gave me my Markov-chain-like progression list, I randomly generated a some test tunes. While most of the results were agreeable, some I didn't like, and ended up trimming the offending chords out of my list.
I realized I probably would need to analyze 3 levels of progressions -- a moving window of A-to-B-to-C -- to get better results, or at least better tests. I never finished that step.
But I should point out that I was interested in generating progressions that sounded good to me. Ultimately I want to make tunes that I like even if I use a computer to help. That means I can use manual adjustments and/or backtracking the chain generation when needed. Therefore, going 3 levels was probably overkill. Leaving it at 2 gives me "happy accidents" anyhow.
Any chance you’ll open source your approach?
As far as open-sourcing it, I'd have to clean it up to be presentable to humans-that-are-not-me, being it's a ball of ad-hoc experiments.
Have you ever considered analyzing the hit progressions thru time, to see if and how our tastes change?
Also, using a self-generated formula that a hit song is probably 25% root progression, 25% melody over that progression (the hooks), and 50% lyrical content...I wonder if some attempt at integrating all three of those items could yield fruitful, interesting results.
Great work, BTW...as a lifelong musician/songwriter, I enjoy sites like yours to help gain a deeper understanding about how professionals manipulate the sonic palette to craft songs and interesting melodies.
I think it's important to be clear that a song is so much more than just the chords. Axis uses a few "tricks" to get things to sound the same:
- strip away everything except the chords & play them all on the same instrument - shift all the songs into the same key - play all the songs at the same tempo
It's a funny exercise that, but it's a bit like saying all cars are the same because they have the same basic structure. Obviously there's a difference between a Camry and a Bugatti.
Amen!
Having grown up on Billy Joel (among others), I actually tend to prefer the sound of IV -> I over V -> I. Glad to see the plagal is equally represented.
Also I wouldn't be surprised if Ab major was the next most common chord. 'I' near a flat-VII and a flat-VI is pretty common. Like Kiss From A Rose from a few years back.
EDM music just takes all these chords and adds a 7th to it
:)
It reminds me of Alfred Bester's 1953 novel The Demolished Man: a murderer pays someone to write a catchy jingle (earworm) for him that can block psychics from reading his murderous thoughts.
for examle; playing a likeable sweet sound when consumer does something good [clicks on ad] or a sour tone when doing "bad" like blocking an ad or popup.
So we constantly get new songs that reuse the same familiarity in different ways, but not literally the same replayed song.
The point here, is that if you look at 1,300 shitty songs, and try to learn about what music is, with those as your example, you are setting yourself up to just make more terrible music.
Maybe a more readily tangible example is: if you were to try and learn about how to create fake dating profile images, but only learned with profile pics of ugly people, then all of your fake dating profiles will probably be unattractive, and they'll never fetch a second look, much less dates.
I mean, this guy cites:
Celine Dion
Taylor Swift
Pharrell Williams
Adele
Bruno Mars
3 Doors Down
Green Day
Lady Gaga
Gorillaz
Linkin Park
Avril Lavigne
Daft Punk
I don't ever want to learn anything about music from this music. It's awful. It all needs to burn. And if you locked me in a room with AI that learned how to create music with these examples as it's template, I'd probably climb the walls and eventually commit suicide. I mean, I'm half way there as it is, and yes, I do blame this kind of music, at least in part.
Well...you seem to know a lot about artists whose music you abhor...I'm curious to see what artists/genres you'd recommend :-)