Genre Suggestion: The App Now Tells You How to Master Your Track
Hit "Measure loudness" — and a chip appears above the genre grid: "💡 Sounds like EDM (strong bass, dense dynamics) [Apply]". That's the new genre suggestion — here's how it's built and why we made it without a neural net.
How it works
The app measures several objective DSP features of your track:
- Five spectral zones (sub-bass, low mids, mids, high mids, air) — the relative energy balance;
- Spectral centroid — the "brightness" of the sound;
- Dynamics — crest factor plus frame-level dynamic range (p95−p10 over 100 ms windows): tells an EDM wall-of-sound from orchestral swells;
- Rhythmic density — spectral-flux peaks per second;
- A speech detector — 3–8 Hz envelope modulation (syllable rhythm) plus energy concentration in the telephone band: podcasts are recognized reliably;
- Stereo width — narrow mono is another vote for podcast.
Genre profiles vote on these features; the best one wins, and the runner-up is shown too. No magic — and that's deliberate.
Why no neural net
Three reasons. Speed: the analysis takes milliseconds and works even on long DJ sets. Predictability: the same track always yields the same suggestion — determinism matters more to us than the word "AI" in a feature list. Explainability: every suggestion arrives with its reasons ("sub-heavy low end", "wide dynamics") — you see why the system decided that way and can deliberately disagree.
A suggestion, not a decision
We intentionally do not auto-apply the genre. The chip shows a hypothesis and its reasons; the Apply button is yours. A genre preset strongly shapes the sound (target LUFS, compression, EQ), and that decision should stay with the artist.
For developers
POST /api/v2/analyze now returns genre_suggestion: style, confidence, reason codes, runner-up. Analysis is free — build your own logic on it: API docs.
Try it: magicmaster.pro/app?lang=en — upload a track and hit "Measure loudness". 1 free basic mastering per UTC day.
Questions and answers
How does the app detect the genre?
With a deterministic DSP heuristic, no neural net: five spectral zones, spectral centroid, crest factor plus frame-level dynamic range, rhythmic density (spectral-flux peaks), a speech detector based on 3–8 Hz envelope modulation, and stereo width. Genre profiles vote on these features; the best one wins.
Is the suggestion applied automatically?
No. It's exactly that — a suggestion: a chip saying 'Sounds like …' with an Apply button and the reasons ('strong bass, dense dynamics'). Choosing the preset is always up to you.
What if the genre is detected incorrectly?
Just pick a preset manually — the suggestion changes nothing until you press Apply. The heuristic honestly shows the reasons for its pick so you can disagree with it deliberately.
Does this work via the API?
Yes: POST /api/v2/analyze returns a genre_suggestion field — style, confidence (0–1), reason codes and a runner-up candidate. Analysis is free.
Upload a track — a finished master in seconds.
Open mastering →AI mastering onlineLUFS analyzer