Chinese national music culture has a long history and contains rich regional characteristics and artistic connotations. Regional cultural characteristics have important value in ethnic piano music creation, but the traditional creation method is less efficient. In this paper, a deep learning-based automatic generation system for ethnic piano music is constructed, which consists of four parts: a score feature extraction layer, a data enhancement module, a Word2Vec-CBOW feature fusion layer and a BiLSTM-CRF fingering generation layer. Specifically, pitch information and velocity information are obtained through score feature extraction, data enhancement is performed by utilizing the symmetry characteristics of the left and right hands, fusion feature vectors are trained using the Word2Vec-CBOW model, and fingering sequence generation is realized by combining the BiLSTM-CRF network. Experimental results show that the system achieves 94.88% accuracy and 92.18% F-value on the MAPS dataset when the task loss weight is 30. The pentatonic scale rate test shows that the pentatonic scale rate of the generated Chinese style piano music in the major key of Celadon reaches 100%, and that of Chrysanthemum Terrace reaches 91.563%. The subjective evaluation experiment in which 40 evaluators evaluated 20 pieces of generated music showed that this paper’s method is superior to the traditional method in terms of coherence and emotional expression. The study shows that the system can effectively integrate regional cultural characteristics and provides a new technical path for ethnic piano music creation.
