Song Dissector

Song Dissector is a single-page web app for taking apart a track to figure out how it works. Drop an audio file in, mark up the structure, loop the parts you want to study, and write down what you hear — everything stays attached to the exact moment it refers to so the song itself becomes the document.

This is what I always wanted Albeton to be able to do, just annotate a song; really any software to do, but I never really found something that could just be used to annotate and take notes on a song, so I ended up building this.

You can play with a demo. It can't save or show you the index, but it does work to dissect a song.

What it does

• Loads any audio file in the browser — MP3, WAV, FLAC, M4A, OGG. The Web Audio API decodes the buffer, a quick autocorrelation pass over the onset envelope estimates the tempo, and the waveform is drawn straight to a canvas at native resolution.
• Three annotation tools layered on the timeline. Sections are coloured bands — drag across the waveform to box one off, then label it (intro, verse, chorus, bridge, solo, outro, anything you want). The loop region is a single indigo band marking the part of the song you want to repeat. Markers are dots in their own lane along the top of the waveform; click to drop one and a notes pane opens for it underneath.
• Plays back with sample-accurate looping. The Space bar plays and pauses, L toggles loop-only playback, and clicks on the waveform seek the playhead. The loop uses the native loopStart/loopEnd properties on AudioBufferSourceNode so the seam is exactly where you set it.
• Notes are bound to whatever is selected. Click a marker, section, or the loop region and the bottom editor loads its notes. Type and the changes save into the project on the next Save. Sections can be renamed; anything can be deleted.

How a project is saved

The Save button (or ⌘S) writes the whole project to the backend: song name, BPM, key, every section/loop/marker with its notes, and on the first save the original audio file itself. Subsequent saves are a JSON-only PUT — the audio is uploaded once and kept on disk.

The home page is a gallery of projects you've saved. Each card shows a deterministic mini-waveform, the song name, BPM and key, counts of sections and markers, whether a loop is set, and how recently it was updated. Click a card to reopen it; the editor pulls the metadata, fetches the audio, and restores the annotations exactly as they were.

Other Stuff

https://github.com/analogpixel/song-dissector has all the code if you are interested in it. This was meant to be deployed to a kubernetes cluster, so there is a bunch of stuff to make that happen in there also.