Classification of musical styles using liquid state machines

作者：Han Ju, Jian‐Xin Xu, Antonius M.J. VanDongen · 年份：2010 · DOI：10.1109/ijcnn.2010.5596470 · 被引用次数：16 · 研究领域：Neural Networks and Reservoir Computing、Advanced Memory and Neural Computing、Neural dynamics and brain function

Music Information Retrieval (MIR) is an interdisciplinary field that facilitates indexing and content-based organization of music databases. Music classification and clustering is one of the major topics in MIR. Music can be defined as `organized sound'. The highly ordered temporal structure of music suggests it should be amendable to analysis by a novel spiking neural network paradigm: the liquid state machine (LSM). Unlike conventional statistical approaches that require the presence of static input data, the LSM has a unique ability to classify music in real-time, due to its dynamics and fading-memory. This paper investigates the performance of an LSM in classifying musical styles (ragtime vs. classical), as well as its ability to distinguish music from note sequences without temporal structure. The results show that the LSM performs admirably in this task.