Use of the entropy of a random process in audio matching tasks

We propose a method to extract a new audio feature which can be used in different audio matching tasks. We use entropy of a random process to estimate the level of information content per chroma of the signal, and we use it as a robust audio feature for different kinds of audio degradations. We have called it entropy-per-chroma. We compare in our experiments entropy-per-chroma against the popular chroma-based feature, since both features look alike in how they are computed and how they show the harmonic and melodic content of an audio-signal. To test the robustness of this new feature, we used audio-fingerprinting approach. Additionally, we tested it on a simple audio-to-audio matching system, using audio of two versions of a same rendition.