This study proposes a model of musical gene expression programming to realize the automatic generation of music of a specific musical style. A music score material library of specific styles of music is established to be used as evolutionary material and neural network training data. In the stage of music evolution, gene expression programming technology is migrated to music composition to form music gene expression programming operations. A mapping mechanism between 16-bit octal numbers and musical scores is proposed to realize the evolution of musical scores in units of measures. In the music evaluation stage, the convolutional neural network evaluation model is trained in a semi-supervised learning manner to evaluate the fitness value of the songs produced by evolution and provide feedback for music evolution. Compared with the music pairs generated by other current studies, the music generated by this research method has obvious characteristics of specific music styles.