Tap “Start Recording” and let NoteSpy listen. We’ll tell you the song playing—fast and accurate.
About NoteSpy
Everything you need to know about how we identify your music
The Journey
From a spark of curiosity to a working music recognition system—here's how NoteSpy came to life.
CuriosityCuriosity
Curiosity
Curiosity to Concept
This project began from wondering how Shazam can recognize a song from a tiny, noisy clip and trying to rebuild that magic for the web. I was fascinated by the idea that a few seconds of humming or background music could be matched against millions of tracks in an instant.
FingerprintingFingerprinting
Fingerprinting
Audio Fingerprinting
Raw audio is transformed into compact fingerprints that capture stable time–frequency patterns, making songs identifiable even with noise and short recordings. These fingerprints are essentially a compressed "signature" of the audio that remains robust across different recording conditions.
KNNKNN
KNN
K-Nearest Neighbors
Each new recording's fingerprint is compared using K-nearest neighbors, finding the closest matches in a high-dimensional fingerprint space. This approach allows us to find similar songs even when the input is imperfect or contains ambient noise.
ScalingScaling
Scaling
Scaling the Search
Ideas from locality-sensitive hashing and vector search keep lookup fast, so matching remains near real-time even as the song catalog grows. These techniques allow us to search through thousands of songs in milliseconds without sacrificing accuracy.
InspirationInspiration
Inspiration
Inspired by Shazam
Research into Shazam's spectrogram constellations and time-aligned hashing guided the design of NoteSpy's feature extraction and matching pipeline. Their pioneering work on audio fingerprinting laid the foundation for modern music recognition systems.