A Fingerprinting-Based Strategy for Musical Genre Similarity

Music information retrieval is a field of signal processing that offers sophisticated tools for generating computer-assisted musical systems. Acoustic fingerprinting is a well-known application of such discipline whose purpose is to extract perceptual features of music samples. The main advantage of implementing audio fingerprinting algorithms is to facilitate the task of matching acoustic samples. For that reason, in this paper, we propose a strategy for measuring musical genre similarity based on the use of audio fingerprints. The notion of musical similarity has been approached employing the use of class-representative fingerprints, defined as an aggregation of the fingerprints of each musical genre. We have also proposed a similarity index defined for the class-representative fingerprints for which we have demonstrated the advantages of using this function as a measure of music resemblance. For the experimental study, we have chosen the GTZAN dataset. The results indicate that our strategy proves to be useful for determining musical similarity between genres. However, future work is required to evaluate and extend it to other domains or datasets.