Soundmango Music Intelligence

What is Soundmango Music Intelligence?

Soundmango Music Intelligence is our own technology for analyzing music and matching each track with the Spotify playlists where it fits. It combines AI-music screening, fingerprint analysis of every track and playlist, track-to-playlist matching, the Playlist Quality Score, and automated campaign workflows.

How well a track fits a playlist and how reliable that playlist is are always judged separately, and a playlist has to pass both before it's matched.

AI-generated music detection

Soundmango only accepts music made by people. We use AI to analyze tracks and run our operations, and we keep that apart from the music we accept.

Generative tools can now produce recordings that sound a lot like conventional music, so we screen every track as part of the quality review, before any curator hears it. A track we judge to be AI-generated is rejected, and the artist gets 100% of their money back.

No current method can tell where every recording came from with complete certainty. Electronic instruments, synthetic sounds, and processed vocals also show up in tracks made by artists, and none of them prove on their own that a track was generated. We read every screening result with those limits in mind.

What is fingerprint analysis?

Every track is a mix of rhythm, tempo, harmony, instruments, vocals, mood, energy, and production. Our fingerprint analysis reads all of that and turns it into a detailed musical profile of each track and each Spotify playlist, so we can see how well a new track fits a playlist before any curator hears it.

Genre labels alone miss most of this. Two indie pop tracks can sound nothing alike, and a downtempo electronic track can sit perfectly next to an alternative R&B one.

Our fingerprint framework uses more than 20 musical and matching factors. Every track gets its own track fingerprint. We then analyze every track on a playlist and combine their fingerprints into one playlist fingerprint. Comparing the two shows how closely a new track relates to the music already on the playlist.

A musical fingerprint differs from the audio fingerprint used for audio recognition, which identifies one particular recording. Two tracks can share no audio at all and still belong on the same playlist.

How a track fingerprint is made

When your track comes in, we analyze its musical characteristics across many dimensions at once and turn them into a structured profile of its musical identity. That profile can then be compared with other tracks and with whole playlists.

Take two tracks that are both called indie pop. One is slow and atmospheric, with electronic textures and restrained vocals. The other has energetic guitars, acoustic drums, and big vocal melodies. They share a genre, and their fingerprints look very different.

It works the other way too. A downtempo electronic track and an alternative R&B track can share their energy level, atmospheric production, light percussion, and vocal style. Their fingerprints pick up on that, even though their genre labels differ.

These relationships decide whether two tracks sit naturally on the same playlist.

A profile in many dimensions

A track fingerprint combines many characteristics into one profile. Some are measurable properties, like tempo or duration. Others are more complex musical qualities, like mood, vocal character, genre influences, and sonic texture. A few draw on metadata or on relationships between tracks, in addition to the audio itself.

The result is a consistent musical reference we can use for every playlist comparison that follows.

The 12 factors behind a track fingerprint

These are the 12 factors we describe publicly. Each one covers a different side of a track's musical identity, and together with our proprietary factors they make up the full fingerprint.

1. Genres and subgenres

Genre gives a first idea of a track's style, but broad genres hold very different music. Electronic covers ambient, downtempo, house, techno, synth-pop, and much more, and rock, hip-hop, jazz, and pop are just as wide.

We look at genres and subgenres to place a track more precisely, and we detect them from the audio, so you never have to pick them yourself. A track can draw on several genres at once. One that blends indie pop, alternative R&B, and electronic production can have a lot in common with tracks from all three.

Genre gives the analysis its context. The other factors tell tracks apart within a genre and connect them across genres.

2. Acoustic or electronic sound

A track built on acoustic guitar, natural percussion, and organic textures sounds different from one built on synths and programmed beats, and many tracks combine both.

We include where a track sits between acoustic and electronic in its fingerprint. That separates tracks that share a genre but differ in production style. An acoustic version of an electronic track keeps the melody and harmony of the original and still has a very different sound.

3. Tempo and key

Tempo is the pace of a track, measured in beats per minute (BPM). Two tracks at the same BPM can still have a different groove, rhythmic emphasis, and feel.

Musical key and other tonal characteristics describe the harmonic side of a track.

Both are part of the fingerprint, and both work alongside the other factors. Matching BPM or matching keys alone tell you little about whether two tracks belong together, since rhythm, mood, and instruments shape how they feel.

4. Mood

Mood is the emotional character of a track: melancholic, uplifting, calm, tense, dreamy, euphoric, or reflective.

It matters most on playlists built around a listening experience. A playlist for relaxing can hold tracks from several genres that share a calm, atmospheric mood. A playlist of uplifting music can mix instruments and styles and still keep one clear emotional direction.

A track can carry several moods at once, and the fingerprint reflects that mix.

5. Energy

Energy is how intense and driving a track feels. Rhythm, dynamics, instruments, and production all shape it.

A fast track can feel relaxed, and a slow one can hit hard through powerful instruments and a dynamic performance. That's why we measure energy on its own. It separates tracks that are alike in other ways but give the listener a different experience, and it matters for playlists built to hold one level of intensity.

6. Vocal character

Vocals shape the identity of many tracks. A soft, airy voice gives a different impression than a powerful, raspy, or aggressive one, even over the same arrangement.

We look at the tone, delivery, texture, and performance style of the vocals. Whether a track has vocals at all matters too, since instrumental tracks often suit different playlists than vocal ones.

7. Language

Some curators run playlists in one language. Others mix tracks from all over the world.

We include the language of the vocals in the fingerprint, and we tell instrumental tracks apart from tracks with singing or spoken words. How much language counts depends on the playlist.

8. Leading instruments

Listeners recognize a style largely by its instruments. A track led by piano, acoustic guitar, or strings sounds different from one built around distorted guitars, synths, or heavy percussion.

We look at the instruments that lead a track and how they sit in the arrangement. Instruments can also connect tracks across genres. A contemporary classical piece and an ambient electronic track can both rest on prominent piano and a sparse arrangement, which gives them a real musical link.

9. Track duration

Some playlists hold short, direct tracks. Others hold long instrumental pieces.

Duration says nothing about quality, and tracks of different lengths can fit together well. A big difference adds useful context, though. A ten-minute ambient track suits a playlist of long atmospheric music better than one made almost entirely of short pop tracks.

10. Explicit lyrics

A track can match a playlist musically and still have lyrics that don't suit what the curator wants. A playlist of clean tracks has different requirements than one that welcomes explicit lyrics.

We include explicit-content information in the fingerprint for that reason. It's a good example of how fit depends on more than how a track sounds.

11. Similar artists

Artists with a similar style often share production, instruments, vocal delivery, or influences. We can use artist similarity to find links that cross genre lines, like two artists in different subgenres whose tracks are closely related.

Artist similarity supports the analysis of the track itself. Artists experiment, and two tracks by the same artist can have very different fingerprints.

12. Similar tracks

We also compare your track with individual tracks. It may share important traits with tracks already on a playlist: rhythm, energy, mood, instruments, vocal style, or a mix of these.

This adds detail to the comparison with the playlist as a whole, and it helps most on playlists that cover several related styles.

More than 10 proprietary factors

The 12 factors above are the ones we describe publicly. On top of them, more than 10 proprietary factors add further detail and catch musical relationships the public factors miss. Some overlap or interact with each other, and others add separate information.

We keep the full list, the methods behind it, the weight of each factor, and how they combine to ourselves. The framework also keeps developing as our methods and data improve.

How a playlist fingerprint is made

A track fingerprint describes one track. A playlist fingerprint describes the musical identity of a whole Spotify playlist, and we build it by analyzing every track on it, never from its name or description alone.

Analyzing every track on the playlist

Picture a playlist with 80 tracks. Each has its own mix of tempo, mood, energy, instruments, vocals, genre influences, and the rest. We fingerprint each one and combine them into one profile of the playlist.

A playlist made mostly of mid-tempo electronic tracks with atmospheric production, restrained energy, and soft vocals gets a fingerprint that reflects exactly that. A playlist of high-energy dance tracks with driving rhythms and big synths gets a very different one. You could call both electronic, and their fingerprints still show how far apart they are.

Combining the fingerprints

At its core, a playlist fingerprint is an aggregate of the tracks on it, a kind of average musical picture of the playlist.

It takes more than averaging the BPM and energy values, though. Some characteristics are numbers. Others are categories, relationships, or more complex data. We use our own methods to combine all of it into one profile.

One identity with room for range

The playlist fingerprint is the reference we compare new tracks with. It shows what kind of music dominates the playlist and which characteristics define it.

A tightly focused playlist gets a narrow, consistent fingerprint. A playlist that spans several related styles gets a broader one. Either way, the playlist has a clear character without every track being alike, and a new track can belong there without resembling every track already on it.

Variation within a playlist

Many playlists hold real variety. A contemporary alternative R&B playlist might mix atmospheric electronic productions, soul-influenced tracks, hip-hop crossovers, and minimal vocal tracks.

The playlist fingerprint describes the whole, and individual tracks sit in different corners of it. A new track can be a strong fit for one corner without being close to the overall average.

Say a playlist is mostly mellow electronic music, with a smaller set of atmospheric alternative pop tracks. A new alternative pop track can belong with that smaller group even though it differs from the playlist's main sound.

That's why we also compare a new track with specific tracks and groups of tracks on the playlist, alongside the playlist fingerprint. It catches fits that the overall profile would miss.

Comparing a track with a playlist

Once your track and the playlists on Soundmango are analyzed, we compare their fingerprints.

A track that shares several important characteristics with a playlist is a strong fit. Differences in the dimensions that matter make it a weaker one. A track with atmospheric electronic production, light percussion, soft vocals, and moderate energy fits a playlist of similar music well. It fits a playlist of aggressive guitars, high energy, and heavy drums much less well.

Fit has many dimensions

Two tracks can have different tempos and still share instruments, mood, and production. Two tracks at the same BPM can have completely different identities. We compare all the dimensions together, which picks up relationships that a simple check of genre or tempo would miss.

Fingerprint similarity feeds the match score

How close the fingerprints are tells us how well your track relates to the music on the playlist. That's the musical foundation of the match score, and the match score also takes in other information about the playlist and its curator.

Beyond fingerprints: more signals in matching

Our matching goes further than the fingerprints. The match score can also reflect information about the playlist and the curator that has nothing to do with how the music sounds. Which signals are available varies from track to track and playlist to playlist.

What a curator has chosen before

A curator's past choices say a lot about their taste. If they've often added tracks with certain characteristics, that's useful context for a new track. Tracks they've passed on can show what suits them less.

A pass on one track doesn't rule out every similar track, though. Taste changes, and every decision has reasons that past data can't fully capture.

Past tracks from the same artist

If a curator has added tracks from an artist before, they've shown interest in that artist's sound, which can count for a new track from the same artist. Links through other tracks or labels can count too, where they exist.

None of this replaces the analysis of the new track. Artists release music in many styles, and a curator's history is one signal among many.

Other proprietary signals

Curator history and artist links are examples, and the full set of matching signals is larger. How much each signal counts depends on the data available and on the track and playlist in question. We keep the full set, the weights, and how they combine to ourselves.

From fingerprints to the match score

We express how well a track fits a playlist as a match score, shown as a percentage. It's the overall result of our matching, with fingerprint analysis as a large part of it alongside musical similarity, context, past patterns, and other signals.

Fingerprint similarity and match score differ

Fingerprint similarity describes how close two musical profiles are. The match score is the overall assessment of fit.

Two playlists can be equally close to your track in sound, and one can still be the better match because of what we know about its curator or the tracks they've added before. So the match scores can differ even when the fingerprints are equally close.

How to read a match score

A higher match score means a closer fit: 92% is a very close fit, 84% a close one, 73% looser, and 48% weak. These numbers are examples to show the idea, not published score bands. A higher score makes a curator more likely to accept the track, since it already fits the playlist's sound.

The score is also a good rule of thumb for how likely the curator is to take on your campaign: at a 92% match, they're about 92% likely to say yes. It says nothing about how much of the audio is identical.

Which playlists qualify

Only tracks and playlists that fit well enough are paired up, so no curator hears a track just because it shares a broad genre with their playlist. A track can qualify for several playlists and miss others, and each curator decides independently. We keep the exact thresholds and weights to ourselves, and we keep refining the model as our methods and data improve.

Across those independent curators, approved tracks have a 99% chance of landing on at least one playlist. Neither the match score nor the 99% promises streams.

Matching accuracy in practice

In our first beta tests, curators accepted 60% of the tracks we matched with their playlists, roughly six out of ten. We only put a track in front of a curator when it closely matched their playlist.

Keeping fingerprints up to date

Spotify playlists change all the time. Curators add tracks, remove old ones, and shift their direction, and each change can move the playlist fingerprint.

New tracks bring new characteristics, and removed tracks take theirs away. A playlist that slowly brings in more energetic music, or drifts toward acoustic tracks, ends up with a different fingerprint over time, and it fits different music than it used to.

We reanalyze playlists automatically as their tracks change, so the fingerprint always comes from what is actually on the playlist. Curators never have to describe their playlist or update genre settings by hand.

Every track and playlist goes through the same framework, so a newly analyzed track is always compared with playlist fingerprints on equal terms. As we improve how we extract, represent, and compare musical features, the comparisons get more precise.

What is the playlist quality score?

The playlist quality score is how we rate the quality, authenticity, and reliability of every Spotify playlist on Soundmango. Each playlist gets a score from 0 to 100, built from many quality signals and the playlist's own history.

The score has three jobs. It helps decide whether a playlist is approved, it keeps quality in check for as long as the playlist is listed, and it sets part of what curators earn from paid marketing campaigns.

Quality and fit are judged separately

The playlist quality score rates the playlist itself. The match score rates how well a track fits the sound of a playlist. A playlist can be a perfect musical fit and still fall short of our quality requirements, so we use both to match tracks with playlists and always judge them independently.

Every playlist is vetted on its own

Every Spotify playlist has to pass our checks before it can be matched with any track. We approve each playlist on its own, so if a curator runs several playlists, one approved playlist doesn't approve the rest. Playlists from the same curator can differ a lot, so we score each one from its own data.

Approval also depends on the playlist's follower count:

• 1,000 followers and up: the playlist can take on paid marketing campaigns.
• 400 – 999 followers: the playlist joins our playlist growth program and gets paid campaigns once it meets both the follower and the quality requirements.

Followers and the playlist quality score are separate checks. Reaching a follower threshold doesn't mean a playlist meets our quality bar.

What the playlist quality score looks at

No single public metric decides the score. How much each signal counts depends on the data available and how it relates to the others.

Follower growth over time

A follower count shows how big an audience is, but on its own it says nothing about whether that audience is real. We look at how the followers have developed over time. A sudden jump isn't automatically suspicious, since playlists grow for good reasons after social media exposure, recommendations, or press. We read patterns over time and never treat a single change as proof.

Playlist activity and upkeep

How a curator looks after a playlist tells us a lot: the tracks added and removed, and whether the activity is steady. A playlist of new releases changes often, while a collection of classics may stay much the same for months, so we judge activity against the kind of playlist it is.

The music on the playlist

We analyze the music on every playlist on Soundmango. Its composition and changes in its content feed into the assessment and help us spot developments that deserve a closer look.

History and consistency

A playlist's history gives context its current numbers can't. Long-term consistency helps us tell an established pattern from something new.

Signs of manipulation

We look for signs of artificial followers, manipulated activity, and other quality problems. An unusual signal doesn't prove manipulation, so we weigh each one against everything else we know about the playlist.

Artist feedback and campaign history

We include artist feedback, whether campaigns were delivered on time, and other observations from completed campaigns. One artist's feedback never decides a score on its own.

How the score affects curator earnings

When two playlists fit a track equally well, the one with the higher playlist quality score is offered more for the campaign. Curators who build strong playlists with real audiences earn more per campaign, and since the score is updated over time, a change in it shows up in future campaign opportunities. We don't publish how much a given score adds to earnings.

Monitoring after approval

A playlist quality score is never final. We keep monitoring every playlist on Soundmango and update its score as new information comes in. Unusual changes can trigger a closer look, and a score can rise or fall as we learn more.

A playlist whose score falls below our quality bar is quarantined. It gets no new campaign opportunities until its score recovers, and then it's listed again. Campaigns it already completed are not affected. Quarantine is a decision about the playlist, never a verdict on the curator, and the curator's other playlists keep their own scores.

Automation and human decisions

Analysis, matching, quality monitoring, campaign tracking, and payouts run automatically. Artists never categorize their track or research curators, and curators never describe every change to their playlist or send an invoice.

Curators decide what goes on their playlists, and our team takes a closer look at quality cases when the automated signals aren't enough. See what that means for you as an artist or as a playlist curator, or read how playlist matching compares with pitching.