Song Genome

v1.3.0
Encoded musical cartography

Turn a song you love into your next song.

Upload a reference. Song Genome measures it, derives its shape, explains what makes it itself — then helps you write your own from that blueprint. Every mark on the page is a measurement.

Methodology v1.3.0 · confidence-scored · your audio stays private
Warming up the analyzer…
Color means one thing only — where a value came from
Measured
signal, layer 1
Derived
computed, layer 2
Interpreted
explained, layer 3
Creative
generated, layer 4
Intensity
peaks & drivers
Uncertain
missing or low

Blue measures · amber derives · wine explains · teal creates · coral intensifies · gray admits uncertainty. Everything else on this page is ink on paper.

The honesty model

Four layers. Confidence you can see without color.

Solid means sure. Dashed means moderate. Dotted means the engine is guessing and says so. Grayscale the whole product and nothing is lost.

L1 · Measured
high confidence
Signal, not opinion
Key G major · 92 BPM · 4/4 · −9.8 LUFS
Direct measurement from the audio and from provider analysis.
L2 · Derived
high confidence
Computed from L1
Energy arc · harmonic rhythm · section contrast
Deterministic transforms with published thresholds.
L3 · Interpreted
moderate
Reasoned, cited
Why it works · Signature · era · sounds-like
Model reasoning, always attached to the measurements behind it.
L4 · Creative
low / exploratory
Yours to break
Blueprints · chord moves · prompts · sketches
Generative output built from your genome, never from thin air.

Click any evidence chip on a claim. The page scrolls to the measurement and flashes it — no claim is more than two interactions from its number.

Live demo · not a screenshot

This is the real track page. Play with it.

Running on the built-in analyzer · no audio, so the playhead sleeps
Analyzing demo track…
The gap

Every songwriter has a reference folder.

Tools today either grade your demo or generate a whole song. Neither helps you learn from a record you love and write your own.

Where we sit

Song Genome sits in the missing seat.

A decoder for songs you admire, so their DNA becomes source material — not a mold. You keep the taste. We do the reverse-engineering.

New in v1.3.0

The genome, drawn properly.

Genome Spiral
One rung per section. Strand separation is energy contrast, thickness is loudness, coral marks the two peaks, wine marks borrowed harmony. It twists at the song's own tempo.
Energy Ribbon 2.0
Measured loudness in blue against derived energy in amber, section boundaries in ink, a live playhead, and click-anywhere-to-seek.
Chord Constellation
The tonic is HOME at the center. Node size is dwell time, path weight is how often you travel it, coral is tension, wine is borrowed.
Chorus Lift Analyzer
Before against after across one section boundary — loudness, energy, harmonic tension and harmonic rhythm — with coral on the delta that actually drives the lift.
Signature claims
Two or three claims naming what makes this recording itself rather than its genre neighbours. Each one points at a concrete trait, each one carries its evidence.
Agents, hands-free
Chord Coach, Producer, Lyricist and Reference Scout — all reading your genome, all citable, and all available by voice when your hands are on the instrument.
The workflow

Reference in. Blueprint out.

01
Isolate
Stems separated in the background — drums, bass, vocals, other. Close the tab if you want.
02
Measure
Key, tempo, chords, energy arc, groove — cross-checked between sources and confidence-scored.
03
Reason
Why it works, the signature, mood and era — every claim with its measurements attached.
04
Create
Transpose, chat with the genome, generate a blueprint, export a Markdown brief.
What we refuse to do
  • No invented data. If section detection returns one whole-track span, the section visuals refuse to draw rather than draw a fake section.
  • No global similarity score. A single percentage flattens four dimensions into one lie, so it appears nowhere in the product.
  • No unmeasured field pretending. It shows an em dash and the reason it is missing.
Rights & privacy
  • Your audio lives in a private bucket scoped to your user ID.
  • Our AI layer only sees extracted numbers — never raw audio.
  • Songs you write from a genome are yours. We claim nothing.
Roadmap

Shipping now, and what's next.

v1.3.0 — shipping now
  • ·Stem separation and background analysis
  • ·Genome Spiral, Energy Ribbon, Chord Constellation
  • ·Chorus Lift analyzer
  • ·Why it works + Signature, evidence-linked
  • ·Agent chat with hands-free voice
  • ·Corrections, library and Markdown export
Next
  • ·Comparison mode across two references (no global score)
  • ·Section-level comparison
  • ·DNA blending
  • ·Atlas cross-catalog matching
  • ·Style memory that follows you across projects
See full roadmap →
Is this a song generator?
No. Generators hand you a finished track. We decode a record you admire so you can write your own.
Do you train on my uploads?
No. Corrections tune thresholds for your library — we never silently retrain models on your audio.
What if a tag is wrong?
Click the pencil and correct it. Your version becomes truth for your library, and the fix propagates to related tags.
A cappella, classical, jazz?
Works — but expect dashed and dotted borders. The engine tells you when it's guessing.
Why no similarity percentage?
Because distance across four musical dimensions is not one number. We show you the dimensions instead.
What does it cost?
Free during beta. Paid tiers land once v1.2 settles.

Load your favorite song. See its genome.

Methodology v1.3.0 · read the full method · the aesthetic playbook