By the numbers
By the numbers
Music theory questions answered with counts from real scores: keys, chords, cadences, tempo, time signatures, vocal range and modulation in real music.
What this section is
Every number in this section was counted from a named, open dataset: real scores, a real music catalogue, real recordings. The main source is the OpenScore Lieder Corpus: 1,462 art songs, read note by note. 1,156 of them also have an automatic chord analysis, and the 2,129 scores of the Mutopia Project, the MusicBrainz catalogue and a sample of 88,816 analysed recordings fill in the rest. Each page gives a short answer first, then the data, then exactly how it was counted and what it cannot show.
- 1,462art songs read from their scores
- 1,156songs with an automatic chord-by-chord analysis
- 2,129Mutopia scores for piano, guitar, voice and more
- 13,669MusicBrainz works with an editor-entered key
Most common keys in music
Which key is most common in music? Six datasets compared: C, F and G major are in every top five, and 26% to 43% of pieces are in a minor key.
Open → ChordsChord progressions in real songs
Chord progressions counted in 1,156 real art songs: V→I 10,374 times, ii–V–I in 498 songs, and after V comes I 73.3% of the time.
Open → HarmonyHow songs end: cadence statistics
How do songs end? In 1,049 art songs, 60.9% of major-key and 46.7% of minor-key songs end V→I in root position, falling from 78% to 24% over time.
Open → TempoTempo markings with real BPM
Allegro and Andante BPM from real scores: Allegro median 100 BPM in art songs and 110 in Mutopia, Andante 66 and 84. Printable tempo markings chart.
Open → KeysGuitar vs piano: which keys do scores really use?
Which keys do guitar and piano scores use? In Mutopia, 59% of 364 guitar scores have sharps; 43% of 484 piano scores have flats. C major leads both.
Open → RhythmTime signature statistics: how common is 4/4?
How common is 4/4? 31.4% of 1,459 art songs and 38.0% of 2,050 Mutopia scores open in it; 3/4, 2/4 and 6/8 follow. Printable chart with real counts.
Open → VoiceVocal ranges in art songs
Vocal range in real songs: the median art-song vocal line runs C♯4 to F♯5 (17 semitones); only 10% stay within an octave. Soprano to bass compared.
Open → HarmonyModulation in real songs
How often do songs modulate? In 1,156 art songs, 79.5% change key for 2+ bars and 27.1% for 8+ bars. The dominant is the commonest target from major.
Open → MelodyHow melodies start and end
Which scale degree does a melody start and end on? In 880 major-key art songs, 42% start on degree 5 and 75% end on the tonic.
Open → MelodyHow music begins: the first notes of 2.0 million real incipits
How do pieces start? In 2,029,719 RISM incipits, 44% of major-key openings begin on the tonic and 35% on the dominant; 30% repeat the first note.
Open →The datasets behind the numbers
Seven open datasets are used in this section. They are collections of mostly Western classical and art music; none is a random sample of all music, and each page says which dataset a number comes from. Licences are those stated by each project (shown below); only totals and shares are published here, not the scores, titles or recordings themselves.
| Dataset | What it is | Size | Licence | Used on |
|---|---|---|---|---|
| OpenScore Lieder Corpus | 1,462 art songs (mostly 19th-century European) transcribed to MusicXML by volunteers; key signatures, time signatures, tempo words, metronome marks, vocal lines and melodies are read directly from the files. | 1,462 songs | CC0 1.0 | keys, chord progressions, cadences, tempo, time signatures, vocal range, modulation, melody, how music begins |
| OpenScore Lieder, automatic harmonic analyses | 1,156 of the songs come with a Roman-numeral analysis produced automatically by AugmentedNet v1.9.1 and not checked by a person; chords, cadences and key changes are counted from those 1,156 files. | 1,156 analyses | CC0 1.0 | chord progressions, cadences, modulation |
| The Mutopia Project | 2,129 free-to-use scores (LilyPond sources and the MIDI files made from them) for piano, guitar, voice, organ, violin and more; keys, time signatures and tempo marks are read from the MIDI or the source header. | 2,129 scores | per piece: public domain, CC BY or CC BY-SA (only counts are shown here) | keys, tempo, guitar vs piano, time signatures |
| MusicBrainz (core data) | The open music database: 2,788,859 works, of which 13,669 carry an editor-entered key (parts and arrangements excluded) and 5,340 have a key in the English title. | 13,669 + 5,340 works | CC0 1.0 | keys |
| Open Opus | A catalogue of 13,014 works by 77 composers; the key is read from the English work title where it has one (3,587 works). | 3,587 titles | public domain (project README) | keys |
| AcousticBrainz | 88,816 recordings analysed by software (Essentia); only the 63,124 keys it was confident about (key strength 0.6 or more) are used, and only as totals. | 63,124 recordings | CC0 1.0 | keys |
| RISM data export (source records) | The international catalogue of music manuscripts and prints: 1,626,011 source records, 1,233,134 of them with a musical incipit; the notes of 2,029,719 incipits were decoded and counted. | 2,029,719 incipits | CC BY 3.0 (credit RISM) | how music begins |
How every number is produced
- Counted by a script, not typed. The pages are generated from the dataset files by one program,
stats_pages_build.py; every figure, ranking and “short answer” is filled in from the counts. - Checked on every build. The program stops if a distribution does not add up, if percentages do not total 100%, or if a value differs from the dataset file. The checks that ran are listed at the bottom of each page.
- Thresholds are stated. Where a result depends on a cut-off (for example how long a passage must last to count as a change of key), the page shows the result at several cut-offs.
- Reading versus estimating. Key signatures, time signatures, notes and tempo marks are read from the files. Major or minor, chords, cadences and key changes are estimates made by software; the pages say so each time.
- Explanations are quoted, not invented. Definitions come from Wikipedia (CC BY-SA 4.0), shown as excerpts with the article, licence and revision date.
Boundaries
- The art songs are a volunteer-transcribed collection of mostly 19th-century European songs, not all songs and not all composers' songs.
- Mutopia and the catalogues lean to what volunteers chose to typeset or document; the recordings are a sample of commercial releases with software-made estimates.
- Nothing here measures popularity, listening habits, or pop, rock, jazz and film music. A common chord in an art song is not a common chord on the radio.
- Counts of automatic analyses (chords, cadences, key changes) were checked against the notes (the labelled chord fits the notes of its bar much better than a random chord) but not against a human-checked reference set.
Licences and credits for everything used on this site are on the credits page.