Jay-Z, Maths and Signals! How to Clone Shazam 🎧
summary
A TypeScript prototype implements a Shazam-like audio recognizer: it reads 44.1 kHz MP3s, converts to mono, splits into 4096-sample chunks, runs a Cooley–Tukey FFT to produce spectrogram frames, and reduces each frame to peaks across six frequency bands to form compact fingerprints. Fingerprints are stored in a database and matched with a simple scoring method; the prototype finds songs from short clips with about a 90% success rate but struggles with noisy, real-time, or stereo inputs.
AI-written summary · about
chapters (10)
- 0:00Shazam intro
- 1:02What is sound?
- 7:20Fourier transform
- 13:20FFT and discrete signals
- 18:12Audio fingerprinting
- 22:02Fingerprint comparison
- 25:12Live demo
- 31:23Shazam differences
- 35:15Technical optimizations
- 36:24Thanks and closing
related talks
51mJay-Z, Maths and Signals! How to Clone Shazam
43mThe Butcherbird Combinator
45mUnderstanding the Mathematical Foundations of Spatial Computing
43mOh, Oh, Oh, It's Magic!
20mVoice fingerprinting & recognition
45mKeynote: How Does Code Sound?
22mSemantic segmentation of tennis match videos
19mYandex Alice: music control with gestures