SPEKTRA

Every track
on one map.

Your library runs deep.
Spektra is a desktop app for DJs. It reads the audio of every track you own and plots your whole library as one map — so you can find music by how it sounds, not just by tags.

Get the free beta
Free while the beta runs. v1.0 is paid.
Pre-1.0 — what works and what doesn’t→
Imports from Mixxx · Rekordbox(local files)·macOS · Windows · Linux
Runs on your machine·No cloud·No subscription·Not generative AI
The Spektra Observatory with a course plotted — a 25-track route threading up through a field of stars plotted by BPM across and Energy up, the Sonic Profile filters at left and the Course track list at right
The Observatory·X = BPM·Y = Energy
The thesis

The structure
is already there.

The library you don’t see

Your library already has a shape.It’s defined by how your music sounds — something a tag was only ever a rough stand-in for. Most of it remains undiscovered. Spektra reads the audio and makes that shape something you can move through.

What you do with it is up to you. Discover. Organize. Plan.

The demo

Watch it
work.

The whole workflow, end to end — read the map, drop anchors, plot a course, audition it, save it.

Recorded on a 1,933-track library·1:43·Silent — no audio track
Read the map

Every star
is a track.

Position, size, and brightness are all measurements from spectral analysis.

The Observatory at rest, zoomed in on 118 to 140 BPM — stars of different sizes and brightness plotted by BPM across and Energy up, with a few in the NO BPM gutter at the far left
Hovering a star shows its cover art, BPM, key, Energy, and its Power, Bass Density and Color bars
Hover any star
Position
BPM across, Energy up. Tracks with no detected tempo sit in the NO BPM gutter on the far left.
Size
Bass Density. Heavier low end reads larger.
Brightness
Color. A track’s tonal character, dark and muted through bright and airy.

What the measurements are

Power, Bass Density and Color come from offline spectral analysis. Energy blends those measurements with an audio embedding. That’s what makes “close” a measurable distance and not a hunch — a useful approximation, not a verdict.

Power
Bass Density
Color

First run

First run reads every file on your drive. The galaxy forms as it goes; embedding is the last pass. None of it leaves your machine.

About what leaves your machine: Except a check for updates, and anything you choose to send. What the app sends

Analysis in progress — Step 1 of 2, Spectral Analysis, 224 of 1,642 ready, about 4 min left
Chart a course

Drop anchors.
Plot a course.

Three moves on the same map. Filter. Anchor. Plot.

01

Narrow

Optional

Six filter groups cut the pool.

Power, Bass Density, Color, BPM, and Times Played are dual-handle ranges drawn over a histogram of your actual library — so you can see what you’re cutting before you cut it. Also filter by your track metadata.

The filter rail expanded — Power, Bass Density and Color dual-handle range sliders over library histograms, with BPM and Times Played sections below
02

Anchor

Click a star, or search by name in Sound Like.

Up to five tracks. Each takes a color and tints its region of the map, so you can see each anchor’s neighborhood before you commit to anything.

Three anchors set — violet, mint and chartreuse chips in the right rail, each showing its match count out of 1,939 in pool, with the canvas tinted by neighborhood
Drag to pan →
Path
Anchor to anchor, in the order you set.
Either
Close to any anchor. Wide net.
Blend
Between all anchors. Common ground.
03

Plot

Set a track count, hit Plot Course.

The route draws across the field with numbered waypoints, the field dims, and the Course list fills the right rail — grouped by anchor and reorderable. Audition tracks.

A course plotted — numbered waypoints connected across the dimmed field, with the Course list in the right rail grouped by anchor
Drag to pan →
Audition

Hear it
before you decide.

Cue any star on the canvas, or any row in a Course, and it plays in the right rail. To hear a whole Playlist, the Recon Deck plays it as a run of clips — 15, 30 or 60 seconds of each track, or all of it — crossfading from one into the next. Full waveform, playhead, Sonic Barcode. You hear where the tracks connect, and where they clash, before you take them to the decks.

The Recon Deck expanded — track 8 of 21 playing, a lens over its 30-second clip on the full waveform, and a strip of mini-waveforms for the rest of the Playlist
Drag to pan →
Save Playlist·Export M3U

Spektra measures, doesn’t generate.
Spektra reads the audio to map what you already own. No models writing your sets, no prompts, nothing invented. The machine learning under the hood is the kind of technique a mastering engineer uses to characterize audio — applied to your crate.

Bought once. Yours forever. Nothing leaves your machine.
No subscription, no cloud, no account, no telemetry. Spektra works on an airplane, and it works in five years whether or not I’m still here.

About what leaves your machine: Except a check for updates, and anything you choose to send. What the app sends

Who makes Spektra?

The playlist is the exit, not the point.
Spektra exports one because that’s the fastest way into the DJ software you already use. What matters is what you turn up on the way there.

Import
Mixxx · Rekordbox (local audio files)
Export
M3U8
Platforms
macOS · Windows · Linux

The track you forgot?
It's already on your drive.

All platforms. Runs on your machine. No subscription, ever.

Get the free beta
Free while the beta runs. v1.0 is paid.
Pre-1.0 — what works and what doesn’t→
macOS·Windows·Linux