Bibliography: Reinhard Sonnleitner, PhD Thesis - 2017 - Audio Identification via Fingerprinting Achieving Robustness to Severe Signal Modifications Sébastien Fenet, PhD Thesis - 2013 - Audio-Fingerprints And Associated Indexing Strategies For The Purpose Of Large-Scale Audio-Identification Avery Li-Chun Wang, Article - 2003 - An Industrial-Strength Audio Search Algorithm
Implementation of a Shazam-like algorithm with audio-fingerprinting and index-searching functions.
Before testing it, you have to:
- change the directory name in main.m, line 10
- create a ./test_songs sub-directory with at least one .mp3 song
- run main.m
The different parts of the program are:
- display the bibliography list
- clear workspace
- set parameters
- set database directory (./test_songs)
- build the database, by computing the audio-fingerprint of each song and saving it in an index
- randomly select a sample of a song in the database directory, and add noise
- divide the sample in overlapping frames (c.f. Fénet's method)
- create the audio-fingerprint of the noisy sample and saving it in an index
- compare both indexes
- analysis of results
Future improvements:
- test the algorithm by studing precision, recall, and computing time
- optimize the algorithm
- Fénet: CQT-based transform
- Sonnleitner: quad-based indexing
- test dereverberation filters