Blog
Since 1986, MusiMap has worked on a shared language capable of describing music consistently, across catalogues, cultures, listeners, and use cases. Taxonomies, lexicologies, and models followed. Understanding came first.
Everyone knows when two songs feel similar.
Very few systems can explain why.
That intuition is where our work begins.
The problem
We recognize mood instantly, whether in a film cue, a radio sequence, a playlist that somehow holds together. We feel the connection before we can name it.
Yet most music systems still describe tracks the way libraries describe books: title, author, date, category.
Useful. Incomplete.
Music carries emotion, memory, culture, identity, and context. A song can feel tender without being slow. A catalogue can contain coherence that genre labels never capture.
That gap is the problem MusiMap has pursued since 1986.
Not by adding more fields to a database. By building a language: a structured, musicologically grounded vocabulary for describing music consistently across millions of tracks, multiple genres, different cultures, and real-world use cases from discovery to broadcasting to sync.
Long before "music AI"
The questions MusiMap works on are not new. Research into music understanding began in 1986, long before the current AI wave. The work already included mood-based discovery, semantic music search, music lexicology, terminology design, synonym handling, and natural language access to music collections: recommendation through meaning, not popularity alone. By 2008, research collaborations including work with LIRMM were already exploring these themes. The vocabulary was still forming. The intent was already there.
The questions were never new. Only the scale changed.
Over time, that language took form: taxonomies, lexicologies, tagging systems, profiling models, metadata structures, and eventually machine learning systems trained to apply the vocabulary at catalogue scale. Each layer was a consequence of the same original intent: to understand music without flattening it.
Machine learning extends reach. It does not define the work. The vocabulary is shaped by musicology, emotional modeling, and expert validation. Technology carries the language further. It does not replace the people who write it.
Describing music requires more than assigning labels. It requires relationships between moods, genres, attributes, contexts, and meanings. MusiMap builds structured lexicologies, not ad hoc keyword sets.
The work begins with how music behaves and how listeners experience it. Engineering follows once the vocabulary is defined, not the other way around.
A shared language only works if it means the same thing across a catalogue, a platform, and a use case. MusiMap optimises for coherence, so “melancholic” or “driving” does not drift with every new dataset.
Metadata tells you who recorded a track and when. MusiMap tries to explain how it feels, what it resembles, and why it belongs beside another song, in terms a curator, algorithm, or supervisor can actually use.
The goal is not a closed system. MusiMap is built as infrastructure: API-first, integrable, designed to sit inside the products and workflows of labels, platforms, broadcasters, and developers who need music to be understood, not merely indexed.
The music industry has standardized how music is stored, licensed, and distributed.
It has never fully standardized how music is understood.
Metadata describes music. MusiMap attempts to explain it.
That distinction matters. Description handles identity (artist, title, release, rights). Explanation handles meaning (mood, feel, similarity, context, resonance. The first is necessary. The second is what makes discovery, programming, recommendation, sync, and creative decisions possible at scale.
MusiMap’s long-term vision is not simply to build products. It is to help the industry develop a deeper, richer, more consistent understanding of music itself: a shared language that survives contact with real catalogues, real listeners, and real commercial constraints.
If that language exists, many things become possible:
Products follow from the language. The language does not follow from the products.
MusiMap is still working on the same problem it started with, not because the answer is easy, but because music deserves better than approximation.
Not a company history, but a record of how one question kept generating new forms of work.
1986
Foundations of Music Intelligence research: emotional modelling, taxonomy design, and music understanding.
2008
The language begins to take shape: taxonomy, terminology, and semantic discovery research.
2010+
Annotation, taxonomy, and platform capabilities mature across real catalogues and real contexts.
Trusted by the Industry
The language tested in production, with leading music companies working at catalogue scale.
Today
The work continues, with the same goal and new means to extend it.
If you work with music at scale, and you have felt the gap between what you hear and what your systems can describe, we would like to hear from you.