Making orchestral construction searchable and audible

What it is

The Music Analyzer tests whether interpretable measurements from orchestral scores can surface musically meaningful passages. I have professionally recorded film music with corresponding MIDI and MusicXML aligned to the recordings by timecode. That makes it possible to analyse the score, retrieve an exact passage, and immediately hear whether the result is useful.

The research question

Reading every note of a dense orchestral score does not automatically explain how music directs attention, persists in memory, accumulates tension, delays resolution, or makes an arrival significant. The project asks whether some of the trained and instinctive understanding used by composers and orchestrators can be surfaced from computer-readable musical evidence.

The current analysis examines evidence related to instrumentation, tempo, meter, activity, register, texture, orchestral mass, dynamics, harmonic and metrical dissonance, energy, tension, and change over time. These feature families have different levels of maturity: some are current outputs, while others remain diagnostic or experimental.

What Score Lens demonstrates

Score Lens is one public interface into the larger Music Analyzer project. Its current sample offers controlled natural-language retrieval across 28 analysed passages from three selected recorded tracks. Search for instruments or musical qualities, choose a result, and hear the recording from the matched timestamp.

The retrieval is deterministic. An optional language-model response can describe the validated matches, but it does not select passages or invent their evidence.

Try Score Lens → Read the development overview →

Measurement is not emotion

The system does not measure emotion as an objective property of music. It measures or records musical conditions that may contribute to perceptions of stability, intensity, tension, momentum, or release. Evocative terms such as peaceful, urgent, or catastrophic are human-readable search language, not emotions detected inside the score.

The wider hypothesis

Generative systems can create impressive musical surfaces and increasingly long continuations. A different question is whether they can treat a motif as a persistent dramatic object: introducing it, passing it around the orchestra, transforming it, withholding it, and returning to it so that a later moment means something because of what came before.

I do not claim to have solved that problem. The present work builds an interpretable way to investigate parts of it across real orchestral music, with human listening and judgement kept inside the method.