Training on your own audio

Training on your own audio

Pink noise teaches you to hear pink noise. The point of importing your own music is that the exercises then hide their changes inside the records you actually mix against.

Importing

Open Training sounds from the bar at the bottom of the home screen.

Add music… opens a file picker - pick as many files as you like. Each one is decoded, cut into eight-second loops, and each loop is sorted by what it measurably is. Progress is shown; it runs on its own thread, so the window stays alive.

Files are copied into the app's own folder. Your originals are never moved, renamed or touched.

How clips are sorted

By measurable character, into five folders:

Character	What lands there	Good for
Percussive	dense transients with gaps between them	compression, delay, distortion
Bass	energy concentrated low	the bottom of the frequency exercises
Mid range	energy where voices and leads sit	general purpose
Bright	cymbals, air, bright synths	the top end
Full range	broad, even, usually wide - a finished mix	the realistic and hardest case

It does not separate stems

It cannot tell a vocal from a mix, and it does not pretend to. That is source separation - a trained-model problem - and a heuristic pretending to do it would mislabel most real music while sounding confident. A lead vocal in a sparse arrangement usually lands in mid range; the same vocal over a dense mix lands in full range, correctly, because that is what the audio is.

Choosing what to train on

Two panes. The left rail is what to train on - pink noise, the built-in synthesized categories, then everything you imported. The right pane is the actual files in whichever one is selected.

• Click a category → the exercises shuffle through its clips, a different one each round. This is the default and the better one: twenty drum loops should be twenty drum loops.
• Click a file → training pins to that clip alone. Click the category again to go back to shuffling.

The footer says which of the two you are in, and the folder path is on screen with a button that opens it.

What is bundled

Five originally-synthesized clips in two always-present categories, so there is a real non-noise option with zero setup. Nothing else - this project does not fetch, bundle or vet the legality of any audio. What you put in your library is yours and your responsibility.

Known limits

• Stereo width always uses pink noise. Clips are downmixed to mono on load, and mono has no side channel to widen.
• Compression and delay are harder to hear on a dense mix than on the synthesized bursts, because both exercises are built around single hits with gaps where an envelope and a repeat are audible. A loop of finished music eats that.