How acoustic timbre search works

When you hear a sound, your brain processes it in a fraction of a second. Bright or dark. Sharp or soft. Sustained or percussive. These aren't just subjective impressions. They correspond to measurable physical properties of the audio signal.

What we measure

SLO extracts a set of acoustic features from every sample in your library. Each one captures a different dimension of what the sound actually is:

• Spectral centroid is the “centre of gravity” of the frequency spectrum. High values mean bright, low values mean dark.
• Transient envelope describes how quickly the sound attacks and decays. A snare has a fast rise. A pad has a slow one.
• Harmonic structure tells you whether the sound is tonal (like a bass note) or noisy (like a hi-hat).
• Spectral flux measures how much the frequency content changes over time. Static sounds score low, evolving textures score high.

Together, these features form a high-dimensional “fingerprint” for each sample. Two sounds with similar fingerprints will sound similar to your ear, no matter what their filenames say.

Similarity as distance

Once every sample has a fingerprint, finding similar sounds becomes a distance calculation. “Find me something like this kick” really just means: find the samples whose fingerprints are closest to this one in the feature space.

That's why SLO can surface a forgotten sample buried six folders deep in a pack you bought three years ago. The filename might be AB_Layer_03_v2_final.wav, but its acoustic fingerprint sits right next to the reference kick you dragged in.

All local, all the time

Every calculation runs on your machine. The feature extraction, the indexing, the similarity search. Nothing leaves your computer. Your samples are your intellectual property, and SLO treats them that way.

The initial scan of a large library takes a few minutes. After that, results come back in milliseconds. The index is cached locally, so rescans are nearly instant.