Every song has a
fingerprint.
Identify a track from ten seconds of its beat. Decode its mood from lyrics and audio with dual engines. Discover what feels the same — while the song’s color washes over the page.
Identify
Name that beat.
Hold your device to the music. A spectral-peak constellation — the same math behind the classic recognizers — is fingerprinted in your browser and matched in milliseconds. No audio ever leaves your device.
Start listeningAnalyze
Two engines, one feeling.
Lyrics flow through a transformer-plus-keyword hybrid; audio through a real MIR pipeline — beat grid, key detection, timbre, valence and arousal. Where words and sound disagree, the combined view shows the tension.
Analyze a songDiscover
Follow the feeling.
Every analysis becomes a 48-dimension sonic fingerprint in a shared mood space. Start anywhere and walk to what feels the same — by sound, not genre tags — then zoom out on the public Mood Atlas.
Explore the space01
Listen
Ten seconds of audio, captured locally. The signal is reduced to its loudest spectral peaks — a constellation unique to the recording, robust to noise.
02
Hash
Peak pairs pack into 24-bit hashes inside a Web Worker. Only these integers travel to the server — the audio itself never does.
03
Align
Thousands of catalog hashes vote on time alignment. A true match is a sharp spike; everything else is noise. Then the mood engines take over.