Series:
Part 1 | Part 2 | Part 3 | Part 4 (this article、final)
What Lies Beyond NotePerformer — Melisma and the Possibility of Music Generation
So far, we have explored how music can be reconstructed from a score, focusing on NotePerformer.
ut where does this approach of “interpreting a score” lead us?
Here, I would like to briefly look at Melisma as one example of what lies beyond this direction.
The Concept of Melisma
Melisma [1] is a system tjat aims to generate audio directly from a musical score using AI.
While traditional sound libraries simply play back sounds, Melisma analyzes the score and generates the performance itself.
In this sense, it shares the same fundamental idea as NotePerformer — reading the score — but appears to take it one step further.
Listening to the available demos, the results already sound quite musical, going beyond a purely experimental stage.
In its newer versions, one can also hear elements of variation and choice in phrasing and dynamics, suggesting a move toward a more generative approach.
[1]: Melisma
The Differences from NotePerformer
NotePerformer provides a stable interpretation based on rules derived from music theory and performance practice.
It is highly reproducible in that essentially the same results can be obtained for the same score.
Melisma, on the other hand, has a mechanism whereby the result is changed by a random seed, and different performances may be generated even for the same score.
This indicates that a stochastic, rather than deterministic, generation process is taking place.
In other words,
- NotePerformer → Interpret the score
- Melisma → Generate music from a score
his is the fundamental difference.
Monophony and Polyphony
An interesting point is that Melisma at this time does not support chords or overlapping notes.
This may be attributed to the approach of generating music as a temporal flow rather than a mere unimplementation.
While the generation of single melody is relatively easy to handle, the generation of polyphony, in which multiple notes, such as chords, are established simultaneously, quickly increases in difficulty both acoustically and musically.
This is because issues such as which notes to bring forward and how to balance them cannot be solved by simple extension.
In practice, it is possible to create a chord-like sound by using multiple tracks, but since each is generated independently, the relationship between the voices is not necessarily taken into account.
This seems to have a somewhat similar structure to early music, in which polyphonic music was established by layering multiple melodies.
From Interpretation to Generation
The sound sources so far are,
- make a sound
- compose a musical performance
This has developed in the direction of
And NotePerformer has reached the stage of "interpreting the score".
Further down the road, we can see the direction of "generating music".
In other words, "score → interpretation → generation.
From the perspective of production
At present, NotePerformer can be said to be more practical, particularly in its handling of polyphony and its ability to produce stable results.
Melisma, on the other hand, has the potential to change the very nature of music production itself.
In particular, the idea that “writing a score directly produces music” represents a significant departure from traditional production workflows.
Conclusion
NotePerformer is a highly refined system that brings music to life through deep interpretation of the score.
Beyond it, we are beginning to see approaches like Melisma, which aim to generate music directly from notation.
At present, this remains largely within the realm of monophony, but the direction is clear, and future developments may reshape the landscape of music creation.
Looking back at music history, we see a progression from monophony to polyphony
In that sense, it may be quite natural that current music generation technologies begin with monophony.
If they eventually move toward polyphony, it may not simply be technological progress, but a process that echoes the history of music itself.
Rather than moving in circles, it advances while transforming its form.
It is not a cycle; it is something that moves forward, changing form.
In other words, music may be evolving in a spiral.
Japanese version is abailable here.
Series Articles