Speech Emotion Recognition Research Based On Attention Mechanism

In recent years,there are more and more academic researches on artificial intelligence technology,and more and more artificial intelligence products appear in the public’s vision.Speech is an important way of human expression,so if artificial intelligence wants to achieve real intelligence,emotion recognition is an indispensable and particularly important direction in research.Speech emotion recognition technology is a bridge for emotional interaction between machines and people.It can be applied to many fields,such as medical treatment,security,transportation and so on.At present,the academic research on speech emotion recognition mainly adopts the method of deep learning.This paper introduces and focuses on the attention mechanism based on emotion cognition,and studies the construction of deep learning network suitable for speech emotion recognition.The main research contents include the following three parts:1.Study how to extract structured emotion information based on attention mechanism.Speech emotional information is dynamic.This paper uses the attention mechanism to calculate the attention weight of emotional features in different dimensions,realize the attention to emotional features,and extract emotional structured information.Specifically,this paper calculates attention from three dimensions:time,frequency and channel.In time,single-layer and multi-layer attention are used to extract the global and local emotional structured information;analyze and improve different methods of calculating attention weight on frequency and channel.Finally,this paper extracts emotional structured information by using attention mechanism,and the UA is improved by 3.48%..2.Propose an relative emotion information extraction method based on attention.Speech emotion is changing relative to the information except emotion in speech.This paper introduces emotional relative information by using multi-task learning and attention mechanism to study and eliminate the negative impact of emotional relative information on emotion recognition.This paper introduces the speaker information and gender information respectively,calculates the attention weight of the information,extracts the emotional relativity information,and pays attention to the emotional information.Finally,the accuracy of emotional recognition is improved through the introduction of emotional relativity information.3.Propose an auxiliary emotion information recognition method based on emotion generation mechanism.This paper summarizes the emotion generation mechanism information,studies the fusion method between this information and emotion information,applies the attention mechanism to realize the auxiliary attention of emotion information by using the emotion generation mechanism information,and finally improves the speech emotion recognition ability of the network.This paper combines emotion theory with practice,applies speech emotion theory information to improve emotion network model,and improves the accuracy of emotion recognition from theory.