Acoustic fingerprint
lastmod 2025-01-21created 2025-01-212 edits
An acoustic fingerprint is a condensed digital summary, a digital fingerprint deterministically generated from an audio signal, that can be used to identify an audio sample or quickly locate similar items in a music database. Practical uses of acoustic fingerprinting include identifying songs, melodies, tunes, or advertisements.
Building a music search system
- Registration: convert existing audio into audio fingerprints (Audio Fingerprint) and store them in a database
- Search: convert the input audio into an audio fingerprint, then compare it with the database to find similar audio
- WAV file: uncompressed PCM format, keeps the original without altering the data
- Sampling: converts an analog signal to digital; the higher the sampling rate, the greater the accuracy
- Spectrogram: a graph that visualizes changes in frequency and amplitude
- Fourier transform: converts a signal to the frequency domain to analyze sound
- Peak: data representing the main frequency components; can remove noise and be compressed
- Audio fingerprint (Audio Fingerprint): generated by hashing peak data; used for storing in and searching the database
- Read a WAV file and generate sampling data
- Apply a Fourier transform to the sampling data to generate a spectrogram
- Extract peaks from the spectrogram
- Hash the peak data to generate audio fingerprints
- Store audio fingerprints in the database, or search with them to find similar audio