Doldur Music data story

What music data
leaves out.

3 collections · 3 different stories

One collection looks across decades, one compares streams with video views, and one maps classical notes. Each reveals something useful—and misses something different.

Read the source notes
28,372English-language songs from 1950–2019
17,967songs compared across streams and views
1.09Mmusical moments marked in MusicNet
10composers in that classical collection

At a glance

Three collections.
Three views of music.

These collections do not begin with the same thing. One counts songs, one pairs songs with videos, and one maps the notes inside classical recordings.

What the three music collections contain and what each is best suited to show
CollectionWhat it containsWhen it looksHow muchWhat it can show
Music Dataset: 1950–2019English-language songs with lyrics and audio features.Release years 1950–201928,372 songs · 5,426 artistsPatterns among the songs included here.
Spotify and YouTubeSpotify songs paired with selected YouTube videos.Numbers collected 7 February 202320,718 entries · 18,862 Spotify songsHow stream and video rankings compare at one moment.
MusicNetClassical recordings with notes and instruments marked in time.Recordings gathered for a 2017 project330 recordings · 34.1 hoursHow computers can learn to recognize notes and instruments.

01 · A long view with gaps

Seventy years.
Not the whole story.

The collection reaches from the 1950s to the 2010s, but the later decades carry far more songs. It follows English-language lyrics across seven genres—one slice of music, not the whole century.

Songs included by decade

28,372 songs · longer bars mean more songs from that decade

  • 1950s1,468
  • 1960s3,409
  • 1970s3,951
  • 1980s4,675
  • 1990s4,457
  • 2000s4,781
  • 2010s5,631
See every decade
Songs and typical audio features by release decade
DecadeSongsShare of collectionTypical acousticnessTypical energy
1950s1,4685.2%0.8520.277
1960s3,40912.0%0.6970.392
1970s3,95113.9%0.3150.514
1980s4,67516.5%0.1500.591
1990s4,45715.7%0.1420.596
2000s4,78116.9%0.1150.684
2010s5,63119.8%0.0890.693
See every genre
Songs and share of the collection by genre
GenreSongsShare of collection
Pop7,04224.8%
Country5,44519.2%
Blues4,60416.2%
Rock4,03414.2%
Jazz3,84513.6%
Reggae2,4988.8%
Hip hop9043.2%

Pop is the largest genre here at 24.8%; hip hop is the smallest at 3.2%. Among these songs, the typical acousticness score falls from 0.852 in the 1950s to 0.089 in the 2010s, while energy rises. That is an interesting shift inside this collection, but it cannot tell us how all music changed. See why in our guides to music dataset bias and genre classification across cultures.

02 · One song, many uploads

A stream is not
a video view.

A collaboration can appear more than once when one Spotify song is paired with several YouTube uploads. We counted each Spotify song once before comparing streams with video views.

How repeated and incomplete Spotify and YouTube entries were handled
What we checkedCountWhy it matters
All song and artist entries20,718A collaboration can create several entries for the same Spotify song.
Different Spotify songs18,862Counting each song once makes the comparison fairer.
Extra entries for repeated songs1,856Leaving them in would give some songs more weight than others.
Songs with more than one entry1,454182 disagree on streams; 1,250 point to videos with different view counts.
Entries without a YouTube match470They cannot be part of a stream-and-video comparison.
Entries without Spotify streams576They cannot be ranked by streams.
Songs compared17,967These songs have usable numbers on both platforms.

Spotify streams and YouTube views

17,967 songs · sorted from lower to higher Spotify streams

Typical Spotify streams and selected YouTube video views from lower- to higher-streamed songs
Stream levelSongsTypical streamsTypical video viewsOfficial video used
Lowest 10%1,7972.4M479.1K74.4%
10–20%1,7979.2M2.6M73.5%
20–30%1,79617.2M4.4M72.6%
30–40%1,79726.9M8.9M74.0%
40–50%1,79740.6M13.1M77.4%
50–60%1,79659.4M17.8M74.8%
60–70%1,79787.2M26.6M78.1%
70–80%1,796135.2M45.6M81.3%
80–90%1,797233.2M76.3M83.9%
Highest 10%1,797562.8M249M89.5%

More-streamed songs usually have more-viewed videos, but the rankings do not move together perfectly. A perfect match would score 1; these rankings score 0.611. One chosen video also cannot capture a song’s total YouTube audience. Read how Spotify streams and YouTube views count different actions.

03 · Detail is not diversity

A million labels.
Ten composers.

MusicNet marks when each note begins, which instrument plays it, and what pitch it holds. That is remarkable detail—but all of it comes from 330 classical recordings by just 10 composers.

Recordings by composer

330 recordings · 34.1 hours · longer bars mean more recordings

  • Beethoven157
  • Bach67
  • Schubert30
  • Brahms24
  • Mozart24
  • Cambini9
  • Dvorak8
  • Faure4
  • Ravel4
  • Haydn3
See every composer
MusicNet recordings and marked musical moments by composer
ComposerRecordingsShare of recordingsMarked musical momentsShare of marked moments
Beethoven15747.6%566,15952.0%
Bach6720.3%62,7765.8%
Schubert309.1%146,57613.5%
Brahms247.3%131,89912.1%
Mozart247.3%75,9307.0%
Cambini92.7%24,8202.3%
Dvorak82.4%31,6052.9%
Faure41.2%22,3492.1%
Ravel41.2%21,1341.9%
Haydn30.9%6,2920.6%
See every ensemble
MusicNet recordings and marked musical moments by ensemble
EnsembleRecordingsShare of recordingsMarked musical momentsShare of marked moments
Solo Piano15647.3%435,15539.9%
String Quartet5717.3%220,31720.2%
Accompanied Violin226.7%97,6409.0%
Solo Cello123.6%10,8761.0%
Wind Quintet92.7%24,8202.3%
Solo Violin92.7%8,8370.8%
Piano Quartet82.4%60,3625.5%
Accompanied Cello72.1%37,5503.4%
Piano Trio72.1%28,8722.6%
Pairs Clarinet-Horn-Bassoon61.8%11,9721.1%
String Sextet51.5%32,1783.0%
Piano Quintet41.2%27,5452.5%
Horn Piano Trio41.2%18,7971.7%
Wind Octet41.2%14,2791.3%
Accompanied Clarinet41.2%10,0490.9%
Violin and Harpsichord41.2%7,4690.7%
Clarinet-Cello-Piano Trio30.9%13,4471.2%
Clarinet Quintet30.9%11,1611.0%
Solo Flute30.9%2,2140.2%
Wind and Strings Octet20.6%11,8871.1%
Viola Quintet10.3%4,1130.4%
See every instrument
MusicNet marked musical moments by instrument and MIDI program
InstrumentMIDI programMarked musical momentsShare of marked moments
Piano1633,59858.2%
Violin41200,46718.4%
Cello4391,1098.4%
Viola4289,2888.2%
Clarinet7224,1502.2%
Bassoon7114,7471.4%
Horn6111,3271.0%
Oboe698,6240.8%
Flute748,3100.8%
Harpsichord74,9140.5%
String bass443,0060.3%

Beethoven alone supplies 47.6% of the recordings, and solo piano supplies 47.3%. The source also estimates that about 4% of its labels are wrong. MusicNet is rich enough to teach note recognition, but too narrow to stand in for music as a whole. Continue with how AI learns music and what ethical training data requires.

04 · What each collection can tell us

Use the right data
for the right question.

Each collection answers a useful question. Trouble starts when we ask it to represent music, listeners, or platforms it never included.

What the three music collections can and cannot tell us
CollectionWhat it showsWhere the story stops
English-language songs, 1950–2019How these songs are spread across decades and audio features.How all music changed or what listeners preferred.
Spotify and YouTubeWhether stream and video rankings move together for these songs.A complete count of platform reach or why a song succeeds.
MusicNetWhich composers, instruments, and notes appear in its classical recordings.How well other genres, cultures, or modern production are represented.

Source notes

Where the numbers
came from.

For the decades view, we used version 3 of Music Dataset: Lyrics and Metadata from 1950 to 2019, updated 31 January 2022 under Attribution 4.0 International (CC BY 4.0), and checked its original study. The researchers began with 82,452 possible songs, then kept those with available English lyrics and cleaned text. The published collection contains 28,372 songs.

For the platform comparison, we used version 2 of Spotify and Youtube, updated 20 March 2023 under CC0: Public Domain. Its numbers were collected on 7 February 2023. When several entries pointed to the same Spotify song, we counted that song once and paired it with the most-viewed official video available—or the most-viewed upload when no official video was listed.

For MusicNet, we used version 1 of MusicNet Dataset, updated 18 February 2021 under CC0: Public Domain, and checked the creator’s dataset record and original paper. We reviewed the recording details and note labels for all 330 recordings without downloading or republishing the music or note-by-note files.

This page keeps only the totals shown here and the source details needed to verify them. Automated checks confirm those numbers whenever the site is tested or built.

Source versions, dates, licenses, files, and SHA-256 checksums used for the analysis
SourceVersion and dateLicense listed by KaggleVerified file checksum (SHA-256)
1950–2019 cataloguev3 · 2022-01-31Attribution 4.0 International (CC BY 4.0)tcc_ceds_music.csv: 046b8e714bb037b5960429a9a4bdfc205530c81e0617ef39a328307c9f9e1173
Spotify and YouTubev2 · 2023-03-20CC0: Public DomainSpotify_Youtube.csv: a08bda951ec1c62a2fbaeb5547141ba23dd7b16347f0e83bd043b1ba9cf168cd
MusicNetv1 · 2021-02-18CC0: Public Domainmusicnet_metadata.csv: 1308d938bafb594e3b0471f2bdda3630da352f881857f265f92114cf398de7ca
label manifest: b942dad583ace4529fb71fb2552a5ca82d8de8343c19e98b674cc69adfb5c57c

Continue exploring

Read beyond
the dataset.

Five guides turn these collection limits into practical questions for listeners, artists, and music-AI builders.

A closer look

Are songs getting
shorter?

See what 8,778 songs suggest about changing track length—and where the answer stops.

Explore song length by era