renared / tipe-audio-fingerprinting

TIPE 2018 : Algorithmes de production et identification d’empreintes de sons musicaux

Audio fingerprinting

TIPE 2018 : Algorithmes de production et identification d’empreintes de sons musicaux

1st program

Inspired by [1].

Identifies peaks in 6 frequency bands of the spectrogram at every time step. Computes the difference between the peaks of a fragment and those of a song and identifies the song that minimizes the difference. Really high computation time (increasing really fast with the number of songs in the database) and too much data stored.

2nd program

Based on Shazam [2], identifies local peaks in the spectrogram, creates pairs of peaks stored as hashes in a SQLite database, identification performed by a SQL query.

3rd program

Based on [3] which uses wavelet decomposition instead of the spectrogram approach, has not been finished.

References

[1] Roy van Rijn : Creating Shazam in Java. : Site consulté régulièrement depuis juin 2017. http://royvanrijn.com/blog/2010/06/creating-shazam-in-java/

[2] Wang Avery : An Industrial Strength Audio Search Algorithm. : p7-13. Ismir. 2003.

[3] Steven S. Lutz : Hokua – A Wavelet Method for Audio Fingerprinting : All Theses and Dissertations. Brigham Young University. 2009.