I analyzed chord progressions in 680k songs

308 点作者 jnord28 天前

37 条评论

Using absolute chord analysis instead of relative chords (i.e. roman numeral analysis) doesn't make sense. As others have noted, the original dataset is flawed because the structure of a song is critical, you cannot omit repeating chords. Programmers/analysts should take more care to understand music theory or the underlying field at hand, before compiling datasets or doing analysis."Most common chord" is mildly interesting, but not really that useful. The most common key, and the most commonly used chords relative to that key (i.e. with roman numeral analysis) would be much more useful and interesting. This would help paint a clearer distinction between e.g. country and jazz, not that "jazz uses Bb major more". Also, anyone with general instrument knowledge would surmise that since Bb and Eb instruments are much more prevalent."If you’re sitting down to write a song, throw a 7th chord in. The ghost of a jazz great will smile on you."7ths don't belong to jazz only, and the average songwriter isn't making data-driven decisions on how to settle on the chord structure for their song.

评论 #43726813 未加载

评论 #43727738 未加载

评论 #43811136 未加载

评论 #43725943 未加载

评论 #43726234 未加载

评论 #43727343 未加载

评论 #43731085 未加载

评论 #43728396 未加载

评论 #43728765 未加载

评论 #43728243 未加载

评论 #43725786 未加载

评论 #43734458 未加载

memset28 天前

The way this analysis, and the original dataset were created, makes no sense. This is, in part, not the author's fault, since the original data [1, 2] is flawed.First, the original data was constructed like this: "...The next step was to format the raw HTML files into the full chord progression of each song, collapsing repeating identical chords into a single chord (’A G G A’ became ’A G A’)..."Already this makes no sense - the fact that a chord is repeated isn't some sort of typo (though maybe it is on UltimateGuitar). For example, a blues might have a progression C7 F7 C7 C7 - the fact that C7 is repeated is part of the blues form. See song 225 from the dataset, which is a blues:A7 D7 A7 D7 A7 E7 D7 A7Should really be:A7 D7 A7 A7 D7 D7 A7 A7 E7 D7 A7 A7With these omissions, it's a lot harder to understand the underlying harmony of these songs.The second problem is that we don't really analyze songs so much by the chords themselves, but the relationships between chords. A next step would be to convert each song from chords to roman numerals so we can understand common patterns of how songs are constructed. Maybe a weekend project.[1] <a href="https://arxiv.org/pdf/2410.22046" rel="nofollow">https://arxiv.org/pdf/2410.22046</a> [2] <a href="https://huggingface.co/datasets/ailsntua/Chordonomicon/blob/main/README.md" rel="nofollow">https://huggingface.co/datasets/ailsntua/Chordonomicon/blob/...</a>

评论 #43725383 未加载

评论 #43725716 未加载

评论 #43725311 未加载

评论 #43724945 未加载

vthommeret28 天前

If you're interested in more relative chord progression analysis, check out Hooktheory (I'm not affiliated but I think love their two books / apps):<a href="https://www.hooktheory.com/theorytab/index" rel="nofollow">https://www.hooktheory.com/theorytab/index</a>It's "just" 32K songs, but you can see the top chord progressions:<a href="https://www.hooktheory.com/theorytab/common-chord-progressions" rel="nofollow">https://www.hooktheory.com/theorytab/common-chord-progressio...</a>And see which songs follow any chord progression you choose (either absolute or relative chords):<a href="https://www.hooktheory.com/trends" rel="nofollow">https://www.hooktheory.com/trends</a>

评论 #43730659 未加载

评论 #43725697 未加载

ben779928 天前

As others have said this is interesting but use of Ultimate Guitar is flawed as the tabs/scores are so bad on that site, very often not even being close to the real chords.On top of being simplified tons and tons of songs get rewritten with a Capo so people can just play G-C-D shapes, if your analysis doesn't look for "Capo" and then transpose all the chords then you end up overrepresenting the key of G and it's chords. Then very often 7th chords, Sus chords, etc.. all get transcribed down to major chords & minor chords due to the beginner focus. Interestingly he doesn't include 6th chords as their own thing.To be fair there are tons of songs that do actually use those chords, so they may still end up coming out as the most popular.I have a grandfathered in lifetime membership to UG that I only had to pay once for. It was cheap so worth it, but I really find the site kind of icky as they are mostly monetizing crowd sourced low quality work and it's very often wrong. And they nerfed their iPad app recently which is really annoying.

评论 #43730910 未加载

评论 #43731050 未加载

评论 #43731876 未加载

评论 #43729856 未加载

cjohnson31828 天前

Listing the "most frequent chord" is a weird analysis, I'm more interested in the "most frequent key", or a transition matrix from one key to another, e.g., if I'm in F, what's the chance I go a fifth up to C, or a fourth down to Bb. Just telling me G is a popular chord doesn't do much.

评论 #43724481 未加载

评论 #43724307 未加载

评论 #43725277 未加载

alexjplant28 天前

Interesting analysis. Some observations:- Ultimate Guitar isn't exactly known for the sterling quality of its transcriptions. Teenage me submitted at least a few tabs that were clearly incorrect that still got 4 and 5 star ratings. Amateur guitarists are also infamously bad at figuring out voicings and extensions so something like a 9 might end up as a maj7 or just a triad. Adult me checks Songsterr first then uses his ear to figure out what's _really_ going on when I run across incorrect parts in the tablature.- Some genres of music like downtuned metal are largely monophonic and instead rely on quick melodic movement or drone-y background guitars to imply harmony. This data set doesn't seem to account for this.- There's no way that power chords only account for single-digit percentages of chords in rock, metal, and punk. There are albums that have been certified Platinum that are 90% power chords (technically power intervals, I suppose).

hirvi7428 天前

I find the analysis interesting in terms of a hobby project, but I'd be careful extrapolating too much out of this. 680k is quite the sample size, but my issue lies within the myopic selection of one instrument and the issues that arise from the platform of Ultimate Guitar.1. I am curious, how many of the 680k songs are unique? It is rather uncommon for massively successful songs to only have one version of tabs out in the wild, so I am curious how many songs individual songs were counted multiple times.2. This analysis only looks at guitar tabs or instrumentations there were transcribed for guitar. Chords can be made with more than just one instrument, thus that missing 7th note could actually be played by another instrument not included in the tabs.3. As music progressed from the pre-jazz era to modern times, it became more common for people to play an instrument, like piano or guitar, while singing at the same time. Obviously there are exceptions to everything, but often times guitar pieces are simplified if the guitarist is also singing for practical reasons.4. Music has also become more accessible as time progressed. It would be hard for an average person to learn the organ or hurdy-gurdy without access to one. It's much easier to acquire and learn piano when it can be a 4 inch thick plastic keyboard on a stand.5. People tend to have a warped concept of the history of music. Pachelbel's Canon in D is by no means a complex song and has stood the test of time. Music through out time has also served different purposes. Hell, go back to Ancient Greece, Gregorian chants, and Medieval music. Those various time periods were not generally fully of complexity either. I would argue such times were generally less complex than modern music.

评论 #43724121 未加载

评论 #43724116 未加载

评论 #43723846 未加载

评论 #43725371 未加载

narrator28 天前

Then there's the most complex pop song of all time: <a href="https://www.youtube.com/watch?v=ZnRxTW8GxT8" rel="nofollow">https://www.youtube.com/watch?v=ZnRxTW8GxT8</a>

评论 #43723929 未加载

评论 #43724277 未加载

评论 #43724221 未加载

divbzero28 天前

Isn’t OP analyzing frequencies of individual chords, not chord progressions?Analyzing individual chords involves counting the frequency of each chord (such as G, C, or D).Analyzing chord progressions would involve counting the frequency of chord pairs (such as D—A or C—G), chord triplets (such as D—A—Bm or C—G—Am), or longer sequences of chords. For an alternative look at the data, you could also normalize chord progressions across key signatures for your analysis (D—A or C—G would both normalize as I—V, D—A—Bm or C—G—Am would both normalize as I—V—vi).

评论 #43725234 未加载

评论 #43725131 未加载

评论 #43724170 未加载

jancsika28 天前

> An “interval” is a combination of two notes.Minor nitpick: it's a "dyad" that is a combination of two notes.An "interval" is the difference between two (or more) pitches. And just as you'd measure the space between your eyebrows using a ruler, you'd measure the interval between middle C and concert A using your ears.The bonus, however, is that our listening apparatus is already quantized to octaves-- if you hear a pitch against a second pitch that's double/quadruple/etc. the frequency of the first, your ear marks this interval as special. It's likely most of you've already used this fact to your advantage; perhaps unwittingly, when someone begins singing "Happy Birthday" outside your normal singing range. (Though most renditions of "Happy Birthday" lend credence to Morpheus' lesson from The Matrix that there's a difference between knowing the path and walking it.) :)

评论 #43725220 未加载

评论 #43724493 未加载

评论 #43724143 未加载

评论 #43723995 未加载

评论 #43729415 未加载

cole-k28 天前

It's an admittedly smaller dataset, but Hook Theory has an analysis that allows you to search by chords (including relative) and look at trends:<a href="https://www.hooktheory.com/theorytab" rel="nofollow">https://www.hooktheory.com/theorytab</a><a href="https://www.hooktheory.com/trends" rel="nofollow">https://www.hooktheory.com/trends</a>It's a weird coincidence to see this post since I only occasionally remember about Hook Theory and binge it, but I remembered earlier this week.Many of you have probably heard the Axis of Awesome four chords song (if not, look it up, it's great), but it's fun doing the same thing with other songs.Like, did you know that you can sing the chorus of Numb by Linkin Park over the chorus of...* I Hate Everything About You by Three Days Grace* Immortals by Fallout Boy* Cheap Thrills by Sia (swung Numb lol)(+ the bridge of The Rock Show by Blink 182)Numb has a pretty common chord progression so I could pick songs with the exact same chords, but there are also some oddly specific finds like this video game (?) song that inexplicably has the same relative chord progression as Hotel California <a href="https://www.hooktheory.com/theorytab/view/zun/reincarnation#Verse" rel="nofollow">https://www.hooktheory.com/theorytab/view/zun/reincarnation#...</a>---I am often surprised how a seemingly simple chord progression has only one result, even when I search by relative chords and ignore extensions and inversions, e.g. <a href="https://www.hooktheory.com/theorytab/chord-search/results?keySelect=any&scaleSelect=any&chordString=i+III+vi+VII&ignoreModifiers=true" rel="nofollow">https://www.hooktheory.com/theorytab/chord-search/results?ke...</a>However when you put that query into the normal search box, it does match a lot more songs, showing that there is a i III _ VII trend, just that i III vi VII is strange (which I guess makes sense). Perhaps my lack of music theory makes it harder to normalize my queries, but it's also possible that (1) there isn't enough data or (2) there is inconsistency in how people annotate the pieces (some songs will have II II II II, for example, following the rhythm, whereas some songs will have just a single II).

评论 #43725180 未加载

throwaway066528 天前

Does this take into account capo position? A G is easy to play so authors might use G to play a Bb for example with a capo to avoid barre chords. Likewise authors will choose simpler chord substitutes to make it easier to play.It's the same with lead sheets / the real book style music books. Performing musicians need to reproduce music quickly so only the triad will be written down even if the musician ends up playing some other extensions.The data is heavily biased towards simplicity. You can make conclusions about the data - but not music as a whole.

mastazi28 天前

I'm surprised that according to the article, in jazz, some chords like D and A, which are mostly found in sharp keys, are more common than chords like Bb and Eb, which are usually found in flat keys.I remember once creating a dataset based on 50 random tunes from the Real Book and sharp keys were less than 20% of the total (based on the key signature at the start of the score) so that graph in the article doesn't seem right.Maybe the discrepancy is because modern jazz fusion tunes are under represented in the Real Book and those are usually more guitar-oriented, so perhaps more likely that the musician would pick a sharp key like D or A. As opposed to straight-ahead jazz were people try to accommodate for sax/trumpet/trombone etc.Or maybe it's because chords like D or A can be dominants in minor keys that are flat keys, e.g. D in the key of Gmin or A in the key of Dmin. - EDIT I just realised that dominants are listed separately so this is not the case.One more thing: according to the article, major triads make up more than 50% of chords in jazz... what? That's certainly wrong, most major chords in jazz are usually maj7th or 6th even when they don't have upper extensions. I think that what they actually meant is "major chords that are not dominants". But they used the label "major triad" instead.

评论 #43727719 未加载

duped28 天前

I think large scale automated harmonic analysis is a worthy endeavor for the purposes of musicological research that could even be applied to pedagogy (deceptively hard problem: identify what piece(s) of music to teach to achieve specific goals for students).But you really need good (and preferably ethical) sources of data to do that, and UltimateGuitar ain't it. You also probably want to engage with some music theorists to normalize the data to give you better analysis and ask better questions than "what is the most common chord."From this analysis I don't think "is music getting simpler" can be answered, and I think the trends are interesting questions to investigate for musicologists but this data set and analysis are too flawed to answer them.

anigbrowl28 天前

One thing that jumped out at me was the data point suggesting there are very few power chords in electronic music. But in fact, they're ubiquitous because it's easy to make a power chord in a single note, by tuning oscillators a 5th apart. Any synth with 2 or more oscillators comes with a bunch of 5th patches (or patch sheets if it's all analog). It's one of the first synthesis techniques people learn to make thick-soundings patches.Also the whole idea of doing the analysis based on absolute rather than relative notes makes little sense to me as a musician, though perhaps that's because I didn't start with guitar or a tuned instrument like a trumpet.

notfed28 天前

This seems to be an analysis of chords used, not chord progressions?

评论 #43724002 未加载

teleforce27 天前

Fun facts you can use circle of fifths as references or cheat sheet for good Chord Progression [1]:"Chord progressions also often move between chords whose roots are related by perfect fifth, making the circle of fifths useful in illustrating the "harmonic distance" between chords."It'll be very interesting to analyse the available songs data to find chords that follow circle of fifths.By cross-reference patterns with the circle of fifths, we might just end up with the LLM equivalent of data-driven musical composer that's capable of generating harmonically pleasing, genre-aware, even hit songs chord progressions.[1] Circle of fifths:<a href="https://en.wikipedia.org/wiki/Circle_of_fifths" rel="nofollow">https://en.wikipedia.org/wiki/Circle_of_fifths</a>

svag28 天前

There is a nice standup sketch from an Australian comedy group, the Axis of Evil, that most of the pop songs are using only four chords, <a href="https://www.youtube.com/watch?v=5pidokakU4I" rel="nofollow">https://www.youtube.com/watch?v=5pidokakU4I</a>

zzo38computer28 天前

I do not see the mention of what chord progressions are used. They did mention what chords (according to only the notes, not according to the key) are common, though.I would expect that a full analysis should write the roman numbers (so you will have to know what key it is as well), and might also consider such things as non-chord tones, modulation to other keys, etc. (However, this is not as simple as just putting them into the computer and writing a SQL query or whatever.)What I had seen on television and what I had read, is that I-V-vi-IV chord progression is common in modern music. (There is also i-VI-III-VII, which is the relative minor key than I-V-vi-IV, which is obvious once you realize it.)

notquitethere27 天前

Notwithstanding how these are read, e.g. C can be Cmaj7 or other substitutions, that are understood by the musician and not always written or transcribed accurately, this analysis is akin to counting the number of times a digit 1 through 12 occurs in a financial statement and analyzing all annual reports. Or taking all the fiction books in the library and counting frequencies of characters translated to ascii codes. These are chords not chord progressions. Progressions would be i-iv-v, ii-v-i, and the prevalence of these. An interesting start, but the main meat of this analysis is unexplored.

digbybk28 天前

I’m having a hard time believing power chords only account for 5.8% of chords in metal.

评论 #43729733 未加载

评论 #43727853 未加载

p0w3n3d28 天前

Just changing key from G too F # does not make the song unique if it still follows the same chord schema. It'd be better to identify chords relatively, i.e. IV->I->V->vi

psychoslave28 天前

Oh just when this morning I was wondering how I might find which sequence of chords would cover all songs ever published.That led me to remember of the famous combinatory proof that was published in Redhit to answer which sequence of some television series one should watch to achieve a minimal chain of something.But I was also concerned that finding chord progression for all songs would be the hardest part, compared to any chord progression which can be just generated by anyone.

calibas28 天前

From my experience playing guitar, the average punk or metal song is almost entirely power chords, while this data says power chords are only 5% of chords in both genres...I thought maybe there's types of metal and punk that I don't know about, but Wikipedia, LLMs and guitar tab sites all agree with me. Punk and metal is overwhelmingly power chords, so I don't see how the data comparing chord types can be correct.

评论 #43728167 未加载

评论 #43728827 未加载

aczerepinski28 天前

As a jazz musician I’d estimate that there are at least an order of magnitude more seventh chords than triads in jazz songs. I question a dataset that says there are more triads. Makes me wonder what else is wrong with the data.For instance, G being the most common key in jazz doesn’t ring true. I’d wager that it’s Bb, Eb, F in some order.Maybe all of the songs in this set were simplified for guitar players?

murki28 天前

what I was hoping this article would do (i.e. express this information as roman numeral progression) has been done here <a href="https://www.amitkohli.com/chord-progressions-of-5-000-songs/" rel="nofollow">https://www.amitkohli.com/chord-progressions-of-5-000-songs/</a>

yahoozoo28 天前

Power chords are only 5% in Metal? OK.

osxman28 天前

Ultimate Guitar is not very accurate. That is because for example blues notes are 'bended' on a guitar by pushing up the strings a lot. Many notes are wrongly notated on Ultimate Guitar. It is better to use official sheet music books.

vitaly-pavlenko28 天前

my favorite paper on analyzing chord progressions so far is this seemingly niche comparative study of five metal genres: <a href="https://pure.hud.ac.uk/ws/portalfiles/portal/58172029/Boddington_Jordan_Herbst_Harmonic_structures.pdf" rel="nofollow">https://pure.hud.ac.uk/ws/portalfiles/portal/58172029/Boddin...</a>it shows which chords exactly set black metal, power metal, metalcore, death metal and progressive metal apart(i also try to do my own narrative on chords, but so far there's nothing to write home about: <a href="https://rawl.rocks/" rel="nofollow">https://rawl.rocks/</a>)

1337shadow28 天前

Aren't Ultimate Guitar chord progressions user contributed? They often simplify the actual chords in their tabs.

RajT8828 天前

What a flawed study - Metal music only uses 5.8% power chords?There is no genre more power chord heavy. (pun intended)

RickJWagner28 天前

That’s a really interesting read.I’m shocked at the similarities between country and punk. Did not imagine that!

sayrer27 天前

A lot of them use G or F because that is the biggest bass note you can get in a club system.

kube-system28 天前

I was disappointed to see they use Ultimate Guitar data for this. Back when I played regularly under the study of a professional, I quickly found that the majority of tabs on the site are just plain wrong.Likely some of the “simplicity” observed in the chords here aren’t actually due to the music being simple, but that the user who contributed the tab made their best simplified guess as to what it was.Unfortunately the analysis is probably flawed because of the poor quality of data.

dboreham28 天前

Oh, the data came from Ultimate Guitar? So as every guitarist knows: it's wrong.

alganet28 天前

HN proceeds to give a crowd pseudo-lecture on notation, because notation is important. I think HN misses the point.If you do these analysis using very basic notation knowledge (letters on top of lyrics are chords!), can you discover by yourself what you missed?Many self-taught amateur musicians go through something similar. You play the simple chord chart, then you notice by yourself that it is not enough. You start to understand the instrument, training the ear, and learning beyond the simple charts.Can you do that with data? Possibly. Maybe, as others mentioned, another dataset would be needed. However, to suggest that such dataset needs to be "a better one in notation" seems misguided.

评论 #43729069 未加载

naijaboiler28 天前

says chord progressions, and then just talked about chords without any progressions. disappointing article.