Robust online music identification using spectral entropy in the compressed domain

Audio identification has been an active research field with wide applications for years. However, most of previously reported methods work on the raw audio format in spite of the fact that nowadays compressed format audio, especially MP3 music, has grown into the dominant way to transmit on the Internet. So far, most of the previous methods take advantage of MDCT coefficients or derived energy type of features. As a first attempt, in this paper we propose a novel audio fingerprinting algorithm utilizing compressed-domain spectral entropy as audio features. Such fingerprint exhibits strong robustness against various audio signal distortions such as recompression, noise interference, echo addition, equalization, band-pass filtering, pitch shifting, and moderate time-scale modification etc. In addition, the algorithm for compressed-domain can be applied in Internet of Things (IoT). Experimental results show that in our test database which is composed of 9823 popular songs, a 5s music clip is able to transmit in IoT and identify its original recording, with more than 90% top five precision rate even under the above severe time-frequency audio signal distortions.