Audio content identification by using perceptual hashing

This paper presents a hashing method for automatic song recognition. This technique works by analyzing the signal and taking into account its nature. The goal of automatic recognition is obtained by extracting different features, which are robust to signal processing and distinctive of the signal. They describe in a compact way the signal and can be efficiently stored in a database. The features are analyzed for short frames using a quantization approach, and a parameter for identification is proposed. Moreover we define also a confidence computed on the identification parameter. By combining the two parameters, a better identification is obtained. The algorithm is tested in different situations: compression, cropping, noise addition, subsampling, stereo to mono conversion, etc. The results show that the identification can be performed also using a short excerpt of the song