Speech emotion recognition method based on long short time memory network and convolutional neural network

Abstract

The invention discloses a speech emotion recognition method based on a long short time memory network and a convolutional neural network. According to the method, a speech emotion recognition system based on LSTM (long short time memory network) and CNN (convolutional neural network) is constructed; with a speech sequence as input of the system, the LSTM and the CNN are trained by virtue of a back propagation algorithm and parameters of the networks are optimized, so that an optimized network model is obtained; and by virtue of the network model which is trained, the newly inputted speech sequence is emotionally classified into six emotions, namely being sad, being happy, being disgusted, being fearful, being scared and being neutral. The method, which takes two network models, namely the LSTM and the CNN, into comprehensive consideration, avoids cumbersome artificial selection and characteristic extraction, and the accuracy of emotion recognition is enhanced.

Description

technical field [0001] The invention relates to the fields of image processing and pattern recognition, in particular to a speech emotion recognition method based on long and short-term memory networks and convolutional neural networks. Background technique [0002] In interpersonal communication, there are many ways of information exchange including voice, body language, facial expression, etc. Among them, voice signal is the fastest and most primitive way of communication, and is considered by researchers to be one of the most effective ways to realize human-computer interaction. For nearly half a century, scholars have studied a large number of topics about speech recognition, that is, how to convert speech sequences into text. Despite significant progress in speech recognition, there is still a long way to go in achieving natural human-machine interactions due to the inability of machines to understand the emotional state of the speaker. This has also led to another as...

AI Technical Summary

Problems solved by technology

Among the above three methods, the main disadvantage is that the advantages of each network model cannot be taken into account

For example, the deep belief network can use a one-dimensional sequence as input, but it cannot take advantage of the correlation between the sequence; although the long-short-term memory network can use the correlation between the sequence, but the extracted feature dimension is high; the convolutional neural network The network cannot directly process the speech sequence. It needs to perform Fourier transform on the speech signal first, convert it into a spectrum and use it as input

The traditional speech emotion recognition method has little development prospects in feature extraction and classification, and the existing deep learning-based speech emotion method network is relatively single

Images