Adapting Image Texture Co-occurrence Analysis for Audio Texture Similarity

In this letter, we adapt a well-known technique of image texture analysis (grey-level co-occurrence matrix) to compute similarity between musical audio signals. Grey-level cooccurrence matrices estimate the joint probability of pairs of pixel values separated by a spatial displacement vector. Instead of using pixel grey-level values, we propose to use frame-based audio features, obtained either analytically with scalar descriptors (such as RMS) or by vector-quantization of Mel-Frequency Cepstrum Coefficients. Second, in lieu of spatial displacement, we analyse co-occurrence with temporal lag between consecutive audio frames. We find that the performance of similarity functions based on this representation depends critically on the features and the strategy of vector-quantization, but not on the temporal lag used for co-occurrence. Overall, we find that the co-occurrence analysis of sound textures does not provide any advantage over first-order histograms in the building of similarity measures. This suggests that second-order statistics, in the form considered in this study, are not a factor as crucial for the computational perception of sound textures as it is for image textures. EDICS Category: AEA-AUEA,IMD-ANAL