Understand the song. No score.

Drop a song on the window. Muzika reads the key and the chords, the tempo, the form, the lyrics and the hooks, and shows where each one sits against 68 years of chart hits — on your Mac, with nothing uploaded. It explains; it does not predict.

Mac App Store How it works

One song, read closely: the key with its honest ambiguity, the chords in Roman numerals, the tempo, the form — and the hook it found in the lyrics, with timestamps.

A report: D minor with the runner-up key and the confidence gap, the chord sequence on a time line with Roman numerals, 144 BPM with tempo confidence and beat stability — Key with its runner-up and how sure the estimate is; every chord on a time line; tempo and how steady the beat is.

The form section: a self-similarity heat map, a spectrogram and a loudness curve with a play head, and the hook candidate with the times it occurs — The song as pictures — where it repeats, how loud it gets, what it looks like — and a player, so every moment the report names is one click from being heard.

Against the charts: one row per feature with the song's value, its percentile among hits, and the pre-registered finding in plain words — Where the song sits among the hits of its era, and what the study actually found — with confidence intervals, not a number out of a hundred.

How it works

Muzika is a statistics engine for popular music with a Mac in front of it. Drag in an MP3, M4A, WAV, AIFF or FLAC. Twenty seconds later you have a report in six parts:

Key and chords. The key with its runner-up and the gap between them, so an ambiguous song is called ambiguous. The chord sequence on a time line with Roman numerals, cadences, harmonic rhythm and chord entropy. Chords are estimated by a hidden Markov model and labelled as estimates.
Tempo and rhythm. BPM with the half- and double-time alternative, tempo confidence, beat stability, onset rate and nPVI.
Form and energy. A self-similarity map that shows where the song repeats, a spectrogram, the loudness curve, section count, time of the first repeat, energy arc and section contrast. No chorus is claimed — the engine does not pretend to know.
Lyrics and hooks. Transcribed on your Mac with Apple's speech recognition. Vocabulary, repetition, the I/you ratio, sentiment, and repeated phrases with the times they occur.
Against the charts. One row per feature: your value, its percentile among the hits of the era you choose, and the pre-registered finding in plain words, with its confidence interval. Where the study found nothing, the row says so.
A player. Click a chord, a hook time, a transcript line or anywhere on a chart and the song plays from there.

Export the whole report as a PDF with one click.

What it does not do

No score. A single number out of a hundred is what people want and what misleads them; Muzika shows position and evidence instead
No server and no account. The analysis, the transcription and the report all happen on your Mac; your song never leaves it
The one network request is optional: once a month it can fetch an updated statistical model — a small file of coefficients and percentile tables with no song data in it. Switch it off and the app uses the copy it shipped with
No adverts, no analytics, no in-app purchases

English and Turkish. Requires macOS 26 on Apple silicon.

How it works

What it does not do

Be there on day one