Beyond Tags and Metadata: Better Music Discovery for Record Labels
Metadata describes what a track is, but does not capture how it sounds.
For decades, music discovery has been based on metadata: genre tags, BPM ranges, mood descriptors and other data that can help a song fit within a certain category.
These are also the standard tools labels use to classify and distribute their catalog, and without them, music cannot be indexed or searched properly.
But there’s a problem. Metadata describes what a track is, but does not capture how it sounds, how it feels or where it fits in different contexts. This causes a discrepancy between what the audiences look for, and what music the metadata directs them to.
A different approach is now possible: analysing the audio itself rather than relying on metadata, and from labeling music to actually analyzing it.
Welcome to Resomix: a new way to explore sounds.
- How Music Discovery Currently Works
- The Limitations of Tags
- Why This Matters for Record Labels
- Beyond Metadata with Sonic Analysis
- How Audio-Based Discovery Works
- Benefits for Labels and Rightsholders
- The Resomix Approach
How Music Discovery Currently Works
Metadata
For the past two decades, tracks have been labeled using predefined fields: genre, tempo, mood and instrumentation. This creates a structured database that can be filtered and searched.
It’s an efficient approach, but not adaptable; it depends entirely on how accurately and consistently those labels are applied.
Editorial Playlists
Editorial curation adds a human layer on top of metadata, with playlist editors selecting tracks based on taste and trends. This can bring a lot of visibility, but there’s a limit: there are only so many slots in a playlist, and selection criteria change from playlist to playlist and are not transparent or scalable.
Algorithmic Recommendations
Streaming platforms use algorithms to recommend music based on listening behavior, relying on a mix of metadata and user interaction data. They might help discover adjacent music genres or artists, but are not designed for depth or alternative exploration.
Across all three layers, the pattern is the same: discovery depends on predefined labels and categories.
The Limitations of Tags
Inconsistency
Metadata is applied by humans, and humans interpret music differently.
One label might tag a track as “ambient,” another as “downtempo,” another as “experimental.” They might all be correct, but the lack of consistency creates confusion.
Music metadata reflects opinion as much as it reflects reality, and as we all know, opinion in music is widely subjective.
Lack of Granularity
Tags group tracks into broad categories but might not be able to register differences in mood and atmosphere that are clear to the human ear.
Two tracks labeled “ambient” might differ completely in density, harmonic movement, spatial texture or emotional tone. For a listener, DJ or supervisor, these differences are critical.
Music metadata reflects opinion as much as it reflects reality, and as we all know, opinion in music is widely subjective.
Static vs. Dynamic Music Context
The same track can be appreciated and used differently depending on context:
- Background listening vs. focused listening
- Opening track vs. peak-time DJ set
- Subtle underscore vs. big-budget sync placement
Metadata does not adapt to contexts; once assigned, tags remain fixed. But contexts do change, and so do people's tastes and needs over time.
Why This Matters for Record Labels
These limitations have a direct commercial impact on how a label generates revenues.
- Discovery opportunities are missed. If a track is mislabeled or under-described, it simply doesn’t appear in relevant searches.
- Catalog utilization remains low. Large portions of the catalog are online but practically invisible.
- Exposure and marketing investments focus on new or already successful tracks. Systems built on engagement and tags tend to highlight tracks that are already successful.
Your deep catalog deserves better than sitting there, unnoticed.
Beyond Metadata with Sonic Analysis
To address these gaps, a different approach is required, one focused on the audio itself. This is where PYXIS-1, Resomix's deterministic sonic analysis engine, comes in.
These are the components that make it a better search engine than a metadata-based system:
- Feature extraction: PYXIS-1 measures more than 70 audio characteristics including spectral balance, dynamics, harmonic content and rhythmic behaviour.
- Timbral similarity: comparing the texture and sonic quality of sounds across tracks
- Harmonic and melodic profiling: analyzing pitch relationships, chord structures, and tonal movement
In simple terms, instead of asking “what is this track labeled as?”, the system asks “what does this track actually do, sonically?”, hence changing discovery from a descriptive model to an analytical one.
How Audio-Based Discovery Works
Signal Analysis (Spectral, Rhythmic, Harmonic Features)
The process begins with signal analysis. Audio is broken down into components:
- Spectral features (frequency distribution, brightness, texture)
- Rhythmic features (tempo stability, groove patterns, transient structure)
- Harmonic features (key, chord progression, tonal complexity)
These elements form a detailed representation of how a track behaves over time.
Similarity Matching
Once extracted, these features are encoded into mathematical representations, called “embeddings”.
Tracks with similar embeddings are close to each other in this “multidimensional space”; this is because two tracks may sit in entirely different genres but share near-identical sonic characteristics. This system helps them match.
Query-Based Discovery (Reference Tracks Instead of Tags)
Instead of searching with tags, users can search with reference points:
- “Tracks like this one”
- “Music with this texture or energy profile”
The system retrieves results based on actual audio similarity, not on how tracks were labeled.
Benefits for Labels and Rightsholders
The implications for catalog performance are obvious and immediate:
Better surfacing of niche tracks: Deep catalog releases that were previously invisible can now be matched to specific queries.
Cross-genre discovery: Tracks are no longer confined to assigned categories. They can appear in multiple contexts based on their vibe.
Increased catalog lifespan: Discovery is no longer tied to release cycles. Older tracks continue to surface as long as they remain relevant to user intent.
These tools can help labels surface more of their catalog and create more discovery opportunities.
The Resomix Approach
We know that metadata is still critical, which is why we believe a hybrid approach is the right solution.
Resomix does not replace tags, but rather enhances them with Pixys-1, a powerful and accurate music discovery engine designed to delve into the soul of music.
The future of music discovery is not purely metadata-based, nor purely audio-driven. Hybrid systems, where classification and analysis work together, offer a more complete model.
For record labels, the advantage is clear: audio-based discovery offers better visibility across their catalog, improved revenue stream of existing assets, and a more direct connection between supply and demand.
Interested in seeing how sonic discovery performs on your own catalog? Contact us today and get a private catalogue evaluation!
