Study of MPEG-7 Sound Classification and Retrieval

In this paper, we present a comparison of three audio taxonomy methods for MPEG-7 sound classification. The MPEG-7 sound classification and indexing tools consist of both low-level and high-level description schemes. For the low-level descriptors that we use, low-dimensional features based on spectral basis descriptors are produced in three stages: normalized audio spectrum envelope, principal component analysis, and independent component analysis. High-level description schemes are used thereafter to describe the modeling of audio features, the procedure of audio classification, and retrieval. For classification we test three approaches: the direct approach, the hierarchical approach without hints, and the hierarchical approach with hints. Our experimental results show that the best approach is the hierarchical approach with hints, which results in a classification accuracy of around 99%. The direct approach produces the second best results, and the hierarchical approach without hints the third best results.