Due to the rapid development of modern science and technology, the teaching methods and modes in traditional music have become increasingly obsolete and outdated, and digital teaching resources are the key to the reform of traditional music teaching methods. In this study, after extracting musical features such as range and average energy of music sounds during the teaching process, it is used as input data for the recurrent neural network model. The model is then optimized using a long-term memory artificial neural network and a self-attention mechanism, and the proposed combinatorial neural network CLSA model is used to visualize the music sounds. Subsequently, a digital music teaching platform is constructed based on the theory of multiple intelligences and the proposed model to assist in the reform and innovation of traditional music teaching methods. After research, the CLSA model proposed in this paper shows good performance in separating vocals and accompaniment, which has reached the performance of the current mainstream model. It is also found that after one semester of experimental teaching, the musical creativity of the students in the experimental class is significantly improved compared with the level before the semester (P<0.05), which indicates that the music teaching based on the digital platform has a good effect. This study provides a reference for relevant music teachers so that the level and teaching effect of digital music teaching can be improved and the deep integration of digital teaching resources and music teaching can be realised.