Music Sense Analysis of Bel Canto Audio and Bel Canto Teaching Based on LSTM Mixed Model

Abstract

As we all know, as the soul of music artists, the cultivation of music sense is an indispensable and important part of Bel Canto teaching. Traditional music classroom education lags behind the development of the information age. According to the educational method of Bel Canto teaching, the recognition experiment of Bel Canto audio is carried out, which helps students analyze the content of music sense contained in music works, and teachers cultivate students' music sense and music theory knowledge according to effective results. In order to better let students appreciate the true meaning of music, it is necessary to add online tools to assist Bel Canto teaching. Traditional methods neither teach students in accordance with their aptitude from the actual situation, but use the rapidly developing computer technology to match resources, nor does it seriously cultivate students' ability to appreciate music and perceive emotions. Based on the above problems, this paper starts from the field of deep learning and plans to build a hybrid model related to LSTM. The results of this paper are as follows: (1) The CNN-LSTM model has the highest recognition rate curve, and the recognition rate of some emotions is over 90%; the loss rate tends to be stable at 200 iterations, and the convergence speed is rapid. (2) After preprocessing, the emotion recognition rate is higher, and the average accuracy of audio features extracted based on spectrogram + LLDs in emotion is about 0.7. (3) According to the actual scene application, the best effect of music sense cultivation is to use the model to assist classroom teaching, and the highest score can reach 8.8 points. In addition, the error between the emotional expression identified by the model and the original work is between 0 and 0.5 points, and the emotional expression effect is excellent. (4) The model can also recognize different kinds and times of emotion in 5-minute Bel Canto works. The above experimental results show that the model basically meets the requirements of the subject, and its performance is excellent, but the details need to be optimized.