Fuzzy Audio Similarity Measures Based on Spectrum Histograms and Fluctuation Patterns

Spectrum histograms and fluctuation patterns are representations of audio fragments. By comparing these representations, we can determine the similarity between the corresponding fragments. Traditionally, this is done using the Euclidian distance. We propose fuzzy similarity measures as an alternative. First we introduce some well-known fuzzy similarity measures, together with certain properties that can be desirable or useful in practice. In particular, we present several forms of restrictability, which allow to reduce the computation time in practical applications. Next, we show that fuzzy similarity measures can be used to compare spectrum histograms and fluctuation patterns. Finally, we describe some experimental observations for this fuzzy approach of constructing audio similarity measures.