Research And Implementation On Emotion Analysis Method Of Shaanxi Opera Based On Deep Learning

Shaanxi Opera is a traditional folk opera in China,which has been passed down for thousands of years.At present,due to the single expression of folk art,the younger generation is no longer enthusiastic about traditional culture,thus Shaanxi Opera is now facing the danger of no one can sing or even disappearing.Therefore,to spread the Shaanxi Opera,modern technology elements can be incorporated into Shaanxi Opera performances.For example,combining virtual reality technologies such as VR display or immersive experience with traditional Shaanxi Opera performances.The purpose is to realize diversified performance forms,which will give Shaanxi Opera a new life.In this way,traditional culture can be protected and inherited.The emotions of Shaanxi Opera are complex and diverse,which makes it is very difficult for a computer to automatically understand.Therefore,the problem of how to make the computer automatically recognize the emotions expressed in Shaanxi Opera is quite urgent to be solved.Currently,deep learning technologies are widely used in audio processing related fields.But the existing models are not accurate enough in classifying long-time series data.Therefore,how to improve the model's accuracy in Shaanxi Opera emotion recognition is the main research content of this article.The research progress as follows:(1)A Multi-head Attention Residual Network(MHAtt?ResNet)is proposed.The spectrogram features contain important emotional information,so this article uses spectrogram as the model input.Combines the residual network(ResNet)and multi-head attention mechanism to make the model have stronger information recognition capabilities.The model proposed in this article can effectively avoid the loss of key emotional characteristics.The model's emotion recognition accuracy on the CASIA dataset reached70.53%,which is better than 56.96% of deep convolutional neural networks and 64.38% of multi-layer residual networks.The results show that the MHAtt?ResNet can effectively improve the accuracy of audio emotion classification;(2)A Convolutional Bidirectional Long Short-term Memory Double Attention Network Model(DAtt?CBLSTM)is proposed.Since the MHAtt?ResNet network will cause the Vanishing gradient when dealing with time series problems,while the Bi-directional Long-Short Term Memory(Bi?LSTM)has the ability to process time series information,the MHAtt?ResNet network is improved by.This model combines Bi?LSTM and attention mechanism,which can well recognize timing features,thereby avoiding problems such as vanishing gradients.The DAtt?CBLSTM model has an accuracy of 85.92% in emotion recognition of Shaanxi Opera,which is much higher than 70% before the improvement.It is obvious that the classification accuracy has been improved.(3)Designed and implemented the Shaanxi Opera emotion classification system.The system is developed using the python language,integrating the algorithms and models proposed above.This article also use the Py Qt5 to displays functional modules such as preprocessing,feature extraction and emotion classification of Shaanxi Opera.Effectively combine theory with practical applications.