Audio Similarity Lab

Upload several audio files, inspect beginner-friendly explanations of what each feature means, and compare timbre, melody, and rhythm with a fuller model that separates instrument colour from pitch motion more clearly.

Pair breakdowns

These pair scores are ranked relative to the current uploaded batch, but the overall score now leans more heavily on melody and harmony than timbre so instrumentation changes matter less than the musical idea.

Top matches first
Similarity note

These are experimental browser-side heuristic scores. They are useful for relative comparison, not copyright or plagiarism judgement.

Time, frequency, and decibels

Compare multiple analyzed files on the same graph. Time runs left to right, frequency rises above the centre line, and decibel strength grows below it.

Tracks to compare
Frequency and Decibel Graph
Drag to pan · Pinch or Ctrl/Cmd + wheel to zoom · Hover for point details

Feature dashboard

This section explains one selected file. The score cards are summaries, but the descriptions underneath are there for people who do not already know audio-analysis terms.

Timbre signature
—
A rough fingerprint of tone colour and spectral shape.
RMS energy
—
Average loudness energy across the clip.
Zero-crossing rate
—
Higher values often mean a rougher or brighter texture.
Spectral centroid
—
A rough brightness centre of the spectrum.
Onset density
—
How many attacks or note starts happen per second.
Estimated tempo
—
A rough beat estimate from repeated attacks.

Feature trajectories

These lines show three changing features across time: centroid is the brightness centre of the sound, flux is how much the spectrum changes from one moment to the next, and flatness shows whether the sound is tone-like or noise-like.

MFCC heatmap

MFCC rows are a compact description of tone colour.

13 timbre rows × time

Pitch register heatmap

This map keeps absolute pitch height instead of collapsing everything into only 12 pitch classes.

pitch height × time