Influence of Guided Particle Swarm Optimization in Automatic Music Emotion Recognition: A Comparative Study Using Different ANN Architectures

摘要

The fundamental operations in soft computing and data mining, has given massive contribution in the development of electronic applications and intelligent system which in recent years, deliberately emphasized on the verbal information such as speech and music. This study presents the development of the music emotion recognition (MER) system that is able to classify 4 different emotions using the low level musical features taken from the Malay music dataset. The artificial neural network (ANN) is used throughout this study as a machine classifier incorporated with guided particle swarm optimization (GPSO) for the purpose of data training. The comparative test using different ANN architectures, with and without the GPSO is proposed as to investigate the impact of combining both algorithms towards the system performance. The results simulate that by incorporating GPSO with the ANN, the classification accuracy can be enhanced up to 90% and more. It is also proved that using GPSO instead of using the conventional PSO techniques somehow improved the musical features learning phase and leads to optimal MER performance.