Sonic: what dance records sound like

What we found

What the measurements show about the music and the people around it. Each finding says who it matters to; tap it for the numbers, how we checked and the catch.

Scenes

What the scenes sound like

Top to bottom: every scene, which scenes sound alike, where the records sit, what sets each apart, what each is made of, and how the scenes have changed.

Every leaf in full

Every scene, by family: what sets it apart and which way it is moving; then sixteen more, grouped by how distinct their sound is. Tap one for its page.

SCENE	WHAT SETS IT APART	MOVING
UK bass · joined by weight in the low end
140 / Deep Dubstep / Grime nearest UK Funky / Gqom · differs by sparser drums	heavier low end · sparser drums · less driving in family: heavier low end	becoming more driving and less defined
Drum & Bass nearest Breaks / Breakbeat · differs by sparser drums	sparser drums · its own key centres · more driving in family: more driving · made of its drums	becoming less defined
UK Funky / Gqom nearest Bass House · differs by busier drums	heavier low end · steadier · drier in family: tighter pulse	becoming less driving and less defined
Breaks / Breakbeat nearest Drum & Bass · differs by busier drums	lighter low end · steadier · heavier low end in family: lighter low end · made of its drums	becoming more driving and less defined
Techno · joined by no voice, relentless drums
Hard Techno nearest Techno · differs by Peak Time — more driving	more driving · less voice · drier in family: more driving · made of its drums	becoming more driving
Techno — Raw / Deep nearest Techno · differs by Peak Time — busier drums	busier drums · less voice · more voice in the mix in family: busier drums · made of its drums	becoming more driving and more defined
Techno — Peak Time nearest Techno · differs by Raw / Deep — sparser drums	busier drums · less voice · more voice in the mix in family: less driving	becoming more driving
Groove house · joined by sparse drums, tight pulse
Tech House nearest Techno · differs by Peak Time — more driving	tighter pulse · less voice in family: busier drums	becoming more driving and less defined
Bass House nearest UK Funky / Gqom · differs by sparser drums	sparser drums · less voice · its own key centres in family: sparser drums	becoming more driving and less defined
UK Garage / Speed Garage nearest Breaks / Breakbeat · differs by more shifting	lighter low end · plainer, less voice · less driving in family: looser pulse	becoming more driving
The house continuum · joined by a voice, unhurried drums
Afro House nearest House · differs by busier drums	more voice · busier drums in family: busier drums	becoming more driving and less defined
House nearest Deep House · differs by lighter low end	lighter low end · more voice in family: lighter low end	becoming more driving and less defined
Deep House nearest House · differs by heavier low end	more voice · warmer · more defined colour in family: more voice	becoming more driving and less defined
Melodic House & Techno nearest Techno · differs by Peak Time — more voice	harder attack · busier drums · more shifting in family: less voice	becoming more defined
On their own · joined by nothing joins them
Amapiano nearest Afro House · differs by harder attack	warmer · more voice in the mix · less voice in the mix · made of its drums	becoming less driving
Trance nearest Melodic House & Techno · differs by more driving	more driving · lighter low end · more voice · made of its melody	becoming less driving
Genres that stand as scenes of their own · Sonic names each more than half the time.
Brazilian Funk named right 83 in 100	most often taken for Trap / Future Bass (8 in 100)	1,727 records
African named right 82 in 100	most often taken for Amapiano (4 in 100)	1,711 records
Jackin House named right 72 in 100	most often taken for Funky House (8 in 100)	1,720 records
Latin Electronic named right 69 in 100	most often taken for Afro House (7 in 100)	1,405 records
Funky House named right 62 in 100	most often taken for Jackin House (13 in 100)	1,579 records
Ambient / Experimental named right 61 in 100	most often taken for Downtempo (9 in 100)	1,687 records
Trap / Future Bass named right 55 in 100	most often taken for Dubstep (10 in 100)	1,773 records
Dubstep named right 53 in 100	most often taken for 140 / Deep Dubstep / Grime (16 in 100)	1,678 records
Hard Dance / Hardcore named right 52 in 100	most often taken for Hard Techno (16 in 100)	1,694 records
Genres that overlap a neighbour · Named right 35 to 50 times in 100; each shares much of its sound with the scene it is most often taken for.
Minimal / Deep Tech named right 44 in 100	most often taken for Tech House (18 in 100)	1,712 records
Trance — Raw/Deep/Hypnotic named right 39 in 100	most often taken for Melodic House & Techno (12 in 100)	1,544 records
Nu Disco / Disco named right 37 in 100	most often taken for Funky House (10 in 100)	1,712 records
Electro named right 37 in 100	most often taken for Breaks / Breakbeat / UK Bass (12 in 100)	1,713 records
Genres that are variants of another scene · Sonic rarely tells them from their neighbours; as sounds, they barely stand apart.
Downtempo named right 35 in 100	most often taken for Ambient / Experimental (13 in 100)	1,696 records
Mainstage named right 24 in 100	most often taken for Trance (10 in 100)	1,710 records
Electronica named right 6 in 100	most often taken for Ambient / Experimental (17 in 100)	1,717 records

Drum and bass shows how the levels fit: it leaves the tree first on speed, its drums are sparser, its nearest neighbour is breaks, and it is becoming more defined. None of it contradicts the rest.

Which scenes sound alike

All 36 genres. Each line joins two that Sonic often mistakes for each other, on records by artists it never heard in training; the thicker the line, the more often. Genres that sound alike sit close together. Tap one.

Tap a scene to see what Sonic mistakes it for.

The records, and the averages

Every released record, coloured by family. No family owns a region: a hard techno record can sit exactly where a deep house record does.

The house continuumGroove houseTechnoUK bassOn their own

driving is brighter and more forward, with less sub. defined is how clearly the parts stand apart and how much the sound moves as it plays. Neither uses harmony. Driving is the line everyone reads, but loudness, drums, tempo and sub bass each tell scenes apart better.

Every measure, every scene

Loudness separates the scenes most, then drums, tempo and sub bass. The two directions every map here is drawn on come fifth and below.

Eight measures, chosen so no two say the same thing. Others were measured and dropped: averaged over a scene, they copy one of these.

SCENE	voice	drums	loud	sub	tempo	pulse	drive
Organic House	+0.7
Afro House
House
Trance				-0.7			+0.7
Indie Dance
Deep House
Progressive
Amapiano			-1.4	+0.7	-0.8
Melodic House
Breaks
Drum & Bass					+0.9
UK Garage
Psy-Trance
UK Funky
Bass House		-0.6
Peak Techno
Tech House						+0.7
140 / Dubstep		-0.9		+1.0			-0.7
Raw Techno	-0.6	+1.0	+0.9
Hard Techno	-0.7		+1.0		+0.8		+0.7

The sixteen genres added to the file later join this table once their full measures are built.

more than the fieldlessdeeper shade, further from the field; the strongest are numbered

Down a column, what a measure does; across a row, what a scene is. Amapiano is quiet, slow and heavy; raw techno dense, loud and voiceless; 140 has the heaviest low end and the sparsest drums.

What each scene is made of

Split every record into drums, bass, melody and voice, and find its nearest neighbour on one part alone. How often does that neighbour share its scene? Chance is one in twenty.

Seven scenes are made of one part. Five are made of their drums, amapiano most of all: its drums alone name it eight times more often than chance. Organic house and trance are made of their melody. None is made of its voice.

SCENE	DRUMS	BASS	MELODY	VOCALS
Amapianodrums · 8.4×	42	28	28	28
Techno — Raw/Deepdrums · 6.3×	31	18	10	13
Hard Technodrums · 6.2×	31	20	15	10
Trancemelody · 6.0×	15	11	30	11
Drum & Bassdrums · 4.7×	24	11	11	9
Organic Housemelody · 3.7×	15	13	19	10
Breaks / Breakbeat / UKdrums · 2.8×	14	8	8	8
140 / Deep Dubstep / Grino single part	24	26	16	15
Bass Houseno single part	16	15	14	14
UK Funky / Gqomno single part	15	9	12	8
Progressive Houseno single part	12	10	13	9
Deep Houseno single part	13	10	12	8
House — Classic/Deepno single part	11	9	10	9
Melodic House & Technono single part	12	8	13	10
Afro Houseno single part	11	9	10	12
Techno — Peak Timeno single part	10	9	8	8
Tech Houseno single part	17	13	18	13
Psy-Tranceno single part	15	16	16	12
UK Garage / Speed Garageno single part	13	15	15	11
Indie Danceno single part	10	8	10	8

how often the nearest record on that part alone shares the scene, in percent; the note is the part that names it best

Between 2024 and 2026 the bass held still while the drums, melody and voice grew brighter and came forward. Records sound more driving because everything above the low end brightened, not because they got busier.

Tested against shuffled years, the drums, melody and voice moved; the bass did not.

48,938 records, eight measures per part. This hears how a part sounds, not what it plays. Computed 20 September 2026.

Where two years has taken each scene

One map, three layers: where each scene sits, where two years has taken it, and how much of everything released it accounts for.

Colour is the family. Families don't own regions of the map: what defines them isn't shown here.

The house continuumGroove houseTechnoUK bassOn their own

Each arrow is two years of travel. Almost every scene is heading the same way, towards more driving.

Size is share of everything released. The biggest scenes sit in the middle; the distinctive ones are small.

What is changing

SCENE chart vs releaseshow newtempo

How we know · Privacy

Artists

Look up an artist or a DJ: their sound, where they sit, who they sound like, and who books them.

How we know · Privacy

ready

tap to listen

Two minutes gives Sonic more to measure when it does not know the record.

Method

What this is

15.1 billionvalues of sound Sonic’s model has learned from: eight slices of the spectrogram of each of 104,842 records

35.9 millionmeasurements taken from the audio: 88 for each of 105,848 records, 53 for each of four parts of the 107,402 split into drums, bass, melody and voice, and 3.8 million fingerprints

105,848released records measured

107,402split into their four parts

36genres, over two years

41,830records it recognises by ear

61,637chart positions

38,198DJ plays, in 1,518 sets

4,064artists placed against their scene

5,566loops measured like a record’s parts

What Sonic does with it

Recognisesa record from a phone held up to a speaker

Separatesevery record into drums, bass, melody and voice

Measurestempo, key and 45 sound measures, for the record and each part

Placesits scene, and where it sits within it

Joinsrecords to artists, labels, DJ plays and bookings

What it solves

What goes next, and why it works: the same drums with a new bass, a key that mixes, where each record sits in its scene, and who plays it.

Producer

Which part keeps a track from where it is aimed, compared part by part with any record, and the licensed sounds closest to each part, from any genre: deep house drums can come from a jungle record.

Sample catalogue

Search a catalogue with a record, part by part, and see what it lacks: the sounds in the charts that nothing in the catalogue comes close to.

A&R

What an artist’s music sounds like, where it sits in their scene and who they sound like; and, from a full reading, whether a demo sits inside a label’s sound.

The full reading, the sounds closest to a record’s parts, the comparison and the tools need an access code; everything else here is open. Figures as of 28 September 2026.

What we cannot see

Measurements wobble month to month because different records come out; a change smaller than that wobble is invisible to us.

The bar is what we measured; the red line is the smallest change we could detect. Only loudness range clears its line.

What this cannot see. Stretched, chopped breaks, which drum and bass is built on, are not recognised by our audio matching. We measure how each part sounds, not the notes it plays. Bookings cover Western Europe well and Johannesburg, Lagos and São Paulo thinly. Genres are Beatport’s labels, which are sometimes wrong. And we hear the best-selling part of each genre, through two-minute previews.

About

Signal is built by Ben. Questions, corrections and ideas are welcome: write to Ben.

What it sees. The sound of released records (Beatport’s charts, sales lists and two-minute previews, across 36 genres and two years); Resident Advisor’s event listings; and what DJs actually play, from published tracklists. Loops come from Freesound, under open licences.

What it does not see. Clubs’ own floors, SoundCloud, Bandcamp, TikTok, Spotify’s editorial playlists, radio and pirate stations. Signal describes the trade’s released and booked view of dance music, not the whole culture, and a record being typical of its scene is not the same as it working at 3am.

What we measure

Version 3.0. Forty-five measurements per record, taken from Beatport’s two-minute preview. Whenever we change how we measure, we measure every record again, 2024’s included, so different years are always compared the same way.

The 45 measurements. Thirteen describe the balance of low, middle and high frequencies; thirteen how that balance changes over time; twelve which musical notes are present; and seven how clearly each element stands out. Apart from the first two, overall level and how dark or bright it is, no single measurement matches something a producer would put a name to.

Driving and defined are the two directions in which records differ most, found in the numbers rather than chosen. Driving leans louder and brighter; defined leans more dynamic, with less voice. The names are shorthand for a direction, not a measure of one thing.

The four parts. Every record is split into drums, bass, melody and voice, and each part is measured for loudness, punch, brightness, how often it hits and its share of the whole. Kick patterns and breakdowns are being added.

Not measured. Melody and chords as notes, words, arrangement beyond breakdowns, and anything outside the preview.

Every record becomes numbers. The instrument has been rebuilt twice, and each rebuild kept more of the record.

What passes through the system, from audio to sentence. Every figure is a count from the database.

41,000minutes of audio heard

about 28 days played end to end

137Maudio points stored

so a record can be recognised inside a DJ set

15.7Mmeasurements taken

768 numbers a record at v3, 45 at v1

21,569records measured

each one a preview downloaded, decoded and analysed

18,000records named to their makers

the join that turns a sound into a person

1,316plays matched inside DJ sets

a record chosen by a DJ, with a timestamp

12,380records recognisable by ear

stored so a phone can search them in two thousandths of a second

452scene-months published

one number for how far a scene stands from its own past

bars are logarithmic: each full step is ten times the one below

34,700measurements behind every number published

The reduction is the point. Twenty thousand records become a few hundred numbers a person can hold in their head. The tests below are there to show the reduction keeps what matters.

Colours mean one thing each: acid is the live signal, warm is caution, and cyan, teal, violet, pink and ink mark the five families of scenes.

How we know it works

All of this assumes a difference we measure is a difference in the music. If Beatport had re-encoded its previews, every drift here would be an artefact.

So we fetched 384 previews again and measured them with the same instrument. Nothing had changed, on any number, for records old or new.

And the zero means something: a re-encode moves these numbers three times as far as a year of the drift we report.

Among records that already charted, chart position is unrelated to anything we measure. Whether a record charts at all is a different question, and it has an answer.

Put charted records beside uncharted ones from the same scenes and months, and the sound alone picks the charted one seven times in ten. A coin gets five.

It holds on a week it has never seen. Trained on the first week of September and tested on this week, it picked the charted record just under seven times in ten, the first test where the answer came after the training. It also works on scenes it was never shown: trained on twelve, it still picks out the thirteenth’s charted records more than six times in ten.

What carries it is texture, not anything a listener would name: not tempo, drum density, voice or how driving a record is.

Held lightly. Three weeks of chart data, one of them lost to a broken capture, and the effect has already halved once. It is the only link we have found between sound and success; four attempts that found nothing are recorded too.

Identical measurements mean identical audio. By that test, one record in fifty here is a second copy of another, under a different Beatport ID.

Where every copy is named, almost all share title and artists, and a quarter changed label between copies: reissues and compilations.

Counted among the records we have measured, not all of Beatport. The full list, with names and links, is in our published data.

Claims on the record

Every claim with the date it was made, when it resolves, and how it fared. The misses stay on the board.

claimmaderesolvesstatus

Melodic house drifts furthest each October and is home by December; it will drift again Aug–Oct 2026.A seasonal wobble, called before the season began, not a scene becoming something else.Aug 2026Dec 2026holding

House music has been rebuilt without changing its parts.Tech house and deep house moved furthest while tempo, drums and voice stayed put: the change is in production. The direction holds; the size was overstated.Sep 2026—stands

Promoters book the old sound.Artists making the newest records average 1.4 bookings; those closest to the old sound average 2.6.Sep 2026Dec 2026open

Records with almost no vocal are dying out.The share fell in ten of thirteen scenes.Sep 2026—stands

Some scenes are really moving; others only wobble. Eight of thirteen scenes move further than shuffling their months can explain, deep house and tech house most. House, peak-time techno, raw techno, garage and breaks do not clear that bar yet.
No scene has a turning point you can date. Scenes drift; none shifts sharply enough to pin to a month.
Records got flatter, not louder. Loudness range fell while loudness itself barely moved.
Records are getting less dynamic everywhere.Eighteen of twenty scenes, a small shift each. Measured on previews; a check against full tracks is still owed.Sep 2026—stands

Deep house’s rebuild is over.Made on an instrument that was discarding a quarter of its numbers. The direction may hold; the size did not. Re-opened when the remeasure finishes.Sep 2026Dec 2026open

Scenes changed in vocal level on average.Withdrawn twice. With every number live, vocals have risen in eighteen of twenty scenes, the opposite of what we first published.Sep 2026Sep 2026withdrawn

Twenty-five sound–attention lead–lag pairs.None survived permutation testing at 14 weeks of demand data. Retest at 26 weeks.Sep 2026Dec 2026failed

Shared audio across records means a shared sample library.Once we named the groups, they turned out to be labels reusing their own recordings under different artist names.Sep 2026Sep 2026withdrawn

The scene map shows how close scenes sit.It exaggerated the gap between the most similar scenes. Removed; the pair table is exact.Sep 2026Sep 2026withdrawn

Tally: 4 stand, 1 holding, 2 open, 4 withdrawn or failed.