The dynamic analysis of multi-voice in chorus music is of great significance to deeply understand the connotation and creative skills of music. Traditional analysis methods have limitations such as low efficiency and strong subjectivity. The purpose of this paper is to construct an accurate and effective dynamic analysis scheme for multi-voice parts of chorus music. By combining time series modeling and deep learning (DL) technology, this paper designs a special time series modeling scheme to capture the time series characteristics of audio, builds a DL model framework combining convolutional neural network (CNN) and long-term and short-term memory network (LSTM), and reasonably sets model parameters and optimization strategies. The experiment uses 600 pieces of chorus music to get 8000 pieces of audio after preprocessing, and it is completed in PyTorch framework on a computer equipped with NVIDIA GeForce RTX 3090 GPU. The results show that the accuracy of the model on the test set is 85%, the recall rate is stable at around 82%, and the F1 value is 83%. In the assessment of different chorus music types, the average absolute error (MAE) of simple style-two voices is 0.04, etc. This method effectively improves the accuracy and generalization ability of multi-voice dynamic analysis, and provides a new way for chorus music research.
Fu et al. (Sun,) studied this question.