Research And Design On Emotion Recognition System Based On Deep Learning

Emoitonal perception is the recognition of personal emotion,which is an important part of artificial intelligence.In order to improve the human-computer interaction experience,and to make machine understand human ‘s emotion better,the academic circles have carried on research from the aspects of human voice,expression,movement and so on.The main content of the thesis is emotional perception from the perspective of voice.Deep Learning is the most popular field of artificial intelligence,and have made significant achievements in aspect of speech recognition,image recognition,natural language processing.the rapid development of the Deep Learning domain has produced so many effective modeling methods such as Deep Belief Network DBN,Convolutional Neural Network CNN,Recurrent Neural Network RNN and so on,and how to use these deep learning methods to improve the accuracy rate in speech emotion perception is a new research issue.In thesis,we focus on how to apply these Deep Learning method to improve the accuracy of emotion recognition.Based on the summarization of the theory of traditional emotional emotion recognition,this paper makes a detailed analysis of various models of Deep Learning.And the Deep Learning model is established using Tensor-flow platform.Based on C/S,the speech emotion recognition system on iOS mobile client is designed.The main contents are as follows:1.The author has a detailed research on traditional methods of emotion perception,and the advantages and disadvantages of traditional emotion recognition methods.Traditional methods of emotion recognition mainly use manual feature extraction,but there are so many kinds of manual features without efficient.The most commonly used is MFCC coefficients,but according to the results of speech recognition in recent years,the effect using MFCC is not better than automatic feature learned by using spectrogram from audio,this paper also introduced spectrum of speech as input to go on automatic feature learning.2.The author studys and analyzes the main algorithm of Deep Learning,the advantages and disadvantages of the Deep Learning methods adopted in the current literature,further proposes XNN-SVM model.Besides,the thesis has used the XNN-SVM to build a system prototype based on Tensorflow.The author carries on some contrast experiments on this system prototype,and proves the improvement effect of the algorithm.3.The author designs and implements a dual-end speech emotion-sensing system based on C/S mode,which can be identified locally by both mobile phone and server,and the architectrure help improve the efficient of the algorithm model by adopting the clients' feedback.Besides,300 piece of voice emotion data were collected to test the system in order to verify the engineering practicability of the algorithm.