Apple has a native API that reads the structure of a song
MusicUnderstanding, tried on eight full recordings from a small Swift command line tool.
Apple has a native framework called MusicUnderstanding. Give it an audio file and it returns a map of what is happening in the music. It is available on OS 27 and later.
It finds beats and bars, sections, segments and phrases, and when vocals and instruments are active. It also estimates loudness, key and pace. The results come aligned to the audio, ready to draw on a timeline.
We tried it with a small Swift command line tool on an M2 Pro Mac, running macOS 27.2 and Xcode 27.1. No Apple Music login. Eight full recordings, about 39 minutes of music, took a median of 6.2 seconds each across warm runs. That is a measurement of this setup, not a promise about yours.
That gives you some useful starting points. A music editor could show phrases alongside the waveform. A video editor could snap a cut to the downbeat. A tool for preparing a set could suggest cue points at section boundaries.
You still need audio you can read. This does not unlock protected Apple Music tracks, and vocal activity is not a separated vocal track. It is not a public AutoMix API. We have not checked the results against human annotations, so accuracy is still an open question.
The figure shows 24 seconds of saved Apple output. The curve is its vocal-activity estimate; the thin bands underneath are its vocal regions. Neither is the audio waveform. The animation starts close on a few beats, then pulls back until the bars, phrases and sections come into view.