Music emotional classification and emotional curve fitting based On BP neural network

This paper concentrates on the MIDI music´s emotional classification and precise emotion measurement. The fundamental theory of this research is BP neural networks. Before the mathematic analysis we carried out a raw data gathering experiment as explained detailedly in paper part I. We analyzed the raw data and picked up a set of MIDI files as the training sample. We put forward a method of classification and curve fitting based on 5 BP neural networks as shown in Figure 1. These networks could be divided into two steps. The Step 1 implements a cursory classification. The Step 2 then gives an accurater emotion measurement. We explain the statistics of input layers and output layers in part III. With the outcomes of network Step 2, we fit the emotional curve as illustrated with cases. As shown in the experiments, these methods have achieved ideal effect of the MIDI music´s emotion measurement.