Skip to content
Generative Music Analysis

Turn any song into visual DNA.

MusicDNA turns a track into a 3D sound genome, painted in real time from rhythm, energy, spectrum, and structure. A 2D mode is also available for clean share cards.

No random dataAnalyzed in the browserNo account
๐ŸŽต

No tracks found.

Add an audio file to /public/music and update manifest.json to start.

  1. 1Add an .mp3, .wav, .ogg, or .m4a file to /public/music
  2. 2Add an entry to /public/music/manifest.json
  3. 3Reload the app
  4. 4Select the track and click Analyze

Seven visualization modes

Each mode is driven by real, deterministic analysis.

01

Song Painting

A generative painting the song paints by itself, brush by brush.

RHYTHMENERGYSPECTRUMHARMONY
02

Sound Genome

A DNA-like helix written by the song. Opens in 3D by default with rotate, zoom, auto-rotate, and camera reset. A 2D mode is available for clean share cards. Beat rungs, energy pulses, and section markers are all driven by real analysis.

RHYTHMSTRUCTUREENERGY
03

Waveform Landscape

A layered audio terrain. Amplitude ridges become mountains, RMS energy forms foothills, spectral bands paint the slopes, and beats pin the horizon.

DYNAMICSSTRUCTURE
04

Rhythm Constellation

Onsets become stars on a bar-phase ร— time map. Downbeats anchor the sky, syncopation draws arcs, density forms clouds, sections tint the regions.

MOTIFSRHYTHM
05

Spectral Aurora

Bass, mids, and highs flow as three aurora curtains. A centroid line traces brightness, harmonic peaks glow, and highs scatter as shimmering particles.

TIMBRESPECTRUM
06

Harmonic Portrait

Chroma vector paints a wheel of tonal weight. Layers grow with harmonic density.

TONALITYHARMONY
07

Track City

Sections become districts. Beats become windows. Energy becomes building height.

STRUCTURESONG

See how one song becomes seven images.

How MusicDNA works

Four steps from sound to signature.

  1. 1

    Choose a track

    Pick any audio file from your music folder.

  2. 2

    Decode the audio

    We decode the file with the Web Audio API in your browser.

  3. 3

    Extract real features

    Rhythm, energy, spectrum, harmony, and structure are extracted from the signal.

  4. 4

    Generate visual DNA

    Deterministic visualizations are rendered from the real analysis.

Visualizations are deterministic. The same file always produces the same DNA.

Compare two tracks

Analyze two tracks side by side and compare BPM, energy, brightness, dynamics, structure, mood, and key moments. MusicDNA helps you understand how two pieces of audio differ, locally in your browser, without uploading your files.

A
Track A
Track A
Start from the selected track
vs
B
Track B
Track B
Pick or upload a second track
  • BPM, energy, brightness and dynamics, side by side
  • Normalized 0-100% timeline alignment
  • Real similarities and differences from the audio

What MusicDNA reads

Every visual is rooted in measured signal features.

Waveform

Downsampled peaks and RMS envelope across the track.

Rhythm

Onset strength, beat times, and BPM estimate via autocorrelation.

Spectrum

FFT frames, spectral centroid, flux, and low/mid/high balance.

Harmony

Chroma vector and key estimate via Krumhansl-Schmuckler profiles.

Structure

Section boundaries from spectral and energy novelty.

Dynamics

Loudness curve, dynamic range, loudest and quietest moments.

Not every feature is available for every track. Uncertain values are clearly labelled as estimates.

Why visual DNA matters

What music looks like changes how you listen.

See the structure

Discover where a track builds tension, slows down, and reaches its peak.

Compare sounds

Spot differences between versions, mixes, and tracks at a glance.

Create visual artifacts

Generate cards, posters, and DNA images from the songs you love.

Analyze without guessing

Every metric is derived from the real audio signal, not invented.

04 ยท Trust

About MusicDNA

MusicDNA turns sound into a visual fingerprint, built entirely from real audio analysis on your device.

No account

No sign-up. No tracking. No upload step.

No random data

Every metric is derived from the real signal.

Analyzed in the browser

Decoding and feature extraction happen on this device.

Export on your terms

Share cards and JSON only when you choose to.

All metrics come from analysis of the selected audio file. Estimated values are labelled. If a value cannot be computed reliably, we do not show it as a fact.

Learn more about the projectSee how it works