Free — up to 2 checks per hour per IP

Music genre detector that actually knows the difference between deep house and tech house.

Record any song around you or upload an MP3 — we'll tell you the genre, sub-genre, BPM and mood. Powered by our own audio AI — trained and maintained in-house. Up to 96% accuracy on GTZAN and MagnaTagATune.

last updated 200+ genres & sub-genres~3s analysis
Live detector
Identify the genre

Sign in to scan — try it free

or upload a file
Deep House
sub-genre · Melodic House & Techno · 122 BPM
94% confidenceeuphoricnight-driveA minor
// what makes it accurate

Built on real research, not vibes.

We built and train our own audio model — paired with a 500+ genre taxonomy we curated from years of real-world data. No guesswork, no charts-based shortcuts.

96%

Benchmark accuracy

On GTZAN and MagnaTagATune we hit 91–96% top-1, depending on genre family. We benchmark on GTZAN and MagnaTagATune and report numbers we measured ourselves.

~3 second analysis

Record 10 seconds, get a result in three. Inference runs on our GPU server; your raw audio never gets stored.

🎛

Sub-genres, not buckets

“Electronic” is too broad. We separate Deep House from Tech House, Drum & Bass from Liquid DnB, Phonk from Drift Phonk.

🎧

BPM & key detection

Beat-grid analysis gets you tempo within ±1 BPM and key in 24 classes — useful for DJs prepping a set or producers hunting reference tracks.

🌐

No sign-up, no ads

Three free analyses in the browser to try, then unlimited with the mobile app. We don't run ads or sell your data. Promise in writing on the About page.

📈

Mood vector

12-dimension mood read: energetic, melancholic, hopeful, dark, dreamy, danceable, aggressive… The same data we use to power “find similar tracks” in the app.

// three steps

How it works.

01

Tap the mic, or drop a file.

We need about 10 seconds of audio. The browser asks for mic permission; on file upload we read the buffer locally — your audio doesn't leave your tab until you commit to analyze.

02

Our model reads the audio.

The audio is processed by our proprietary model — trained on millions of labeled tracks across 500+ genre categories. It scores every genre simultaneously and re-ranks with a fine-tuned head trained on curated real-world data.

03

Genre, sub-genre, BPM, mood — in 3s.

You get the top label with a confidence score, the runner-up genres in case it's a hybrid, and the BPM/key/mood breakdown. Save to favorites, share a result link, or analyze another.

// browse the taxonomy

A few of the 200+ genres we know.

Tap any chip to see example tracks our detector found in the wild.

+ 174 more sub-genres
/* how this thing is built */

Our own model. Built for music, not borrowed.

Most genre detectors repurpose general audio embeddings. We took a different path — training a dedicated model on millions of labeled tracks, fine-tuning it specifically for sub-genre granularity. That's why it separates Deep House from Tech House, Drum & Bass from Liquid DnB, Phonk from Drift Phonk. We benchmark on GTZAN and MagnaTagATune and report numbers we measured ourselves.

// questions, mostly real

FAQ.

Shazam matches an audio fingerprint against its catalog of known tracks. If the song isn't in the catalog (a DJ edit, a bootleg, a release you bought on Bandcamp), it gives up. We don't try to identify the song — we listen and tell you what kind of music it is. So a 1996 vinyl rip and a yesterday's SoundCloud upload both get analyzed.
Yes. Three free analyses in the browser, no sign-up. We pay for the GPU server out of mobile-app subscription revenue. If you need more, the iOS/Android app gives you unlimited analyses for the price of a coffee a month.
Anything from a phone-mic recording in a noisy bar to a lossless WAV will work. Accuracy goes up with cleaner audio — at 128 kbps MP3 you'll still hit ~88% on the GTZAN benchmark; at 320 kbps or lossless we're at 94%+.
No. We hold the audio in memory only as long as needed for the embedding pass, then drop it. The result (genre + BPM + mood) is logged to a result ID so you can share it; the source audio isn't.
That's the whole point. Our taxonomy has 200+ leaves and the classifier head was fine-tuned specifically to disambiguate close pairs. You'll get a top-1 sub-genre with confidence and the runner-ups for ambiguous tracks.
Tracks that sit between genres are genuinely ambiguous — a song that's 60% Trap and 40% Phonk will get scored as either depending on which intro you sampled. We show confidence and runner-ups so you can spot when the model is unsure. For really fringe stuff (drone, free jazz, microsound) the taxonomy isn't deep enough yet.
Not yet publicly. If you're a DJ pool, library or B2B partner with a real use case, say hi and we can talk pricing.

Stop calling everything “house”.

Free in your browser. Unlimited in the app.

Get on iOS →Get on Android →
Free Online Music Genre Detector — Genre AI