Audio hashing technique for automatic song identification

We present a novel method for automatic song recognition based on a hashing technique. The goal of the project is to automatically retrieve from a database some information regarding a song without any previous knowledge and without the insertion of additional information. The goal is obtained by analyzing the signal and extracting a robust feature. That feature can be used to make a distinction between different songs but also to identify as similar two versions of the same songs. Moreover we analyze also the error probability and we define a confidence for the obtained match. The obtained results show that the song identification can be done with high confidence even if we use only a little excerpt of the song.