Audio Fingerprinting

Epistemic Review

Sources

• How audio fingerprinting works?
• An Industrial-Strength Audio Search Algorithm - Shazam

Audio fingerprinting is the method used by Shazam and YouTube to identify music from sound. AI methods are also being heavily used for audio fingerprinting and search these days

1. Compression using Huffman Coding. Lossless compression isn’t necessary
2. Hashing ← Compact result, can be quickly searched and sorted with log(N) complexity
3. Audio fingerprinting is converting audio into a compact and quick-searchable format to compare and identify audio
4. Convert to raw .wav file and check for sample rate, bit depth, number of channels
5. Produce a spectrogram
6. Encode the spectrograms for image search. Using locally sensitive hashing and the Jaccard coefficient as a distance metric