Research On Variable Length Speech Emotion Analysis Based On Neural Network

Speech is one of the most important ways for human beings to communicate and understand the world.The information contained in it is second only to vision.With the rapid development of artificial intelligence,man-machine interactive has become one of the hot spots of current researches.The ability of machines to recognize and express emotions has become the goal of researchers.The importance of speech emotion analysis has become increasingly important.In the field of speech sentiment analysis,the current common research is to study the fixed-length speech sentiment analysis,especially the variable-length speech sentiment analysis.Due to the variability of the length of the speech signal itself,this paper mainly studies the speech sentiment analysis of variable-length speech signals..In real life,the length of speech signals is basically variable,so the research in this paper makes a difference in theoretical significance and has a little application prospects.In terms of feature extraction of speech signals,this paper studies common features:spectral features,prosodic features,and sound quality features.At present,most researchers use a combination of multiple features,namely fusion features for experimental research.Since the MFCC can effectively reflect the auditory characteristics of the human ear,the features extracted in this paper are the MFCC features in the commonly used spectral features.The main work of this paper is as follows:(1)The research background and significance of speech sentiment analysis are studied.The research history and current situation of speech sentiment analysis are summarized.The research status of speech emotion analysis of variable length sequences is summarized.(2)Two kinds of processing methods are proposed for the speech sentiment analysis of variable length sequences,and the methods are analyzed to analyze their influence on spectral features and partial prosodic features.(3)The BP neural network model is designed,and the speech sequence is processed by the speech-sense analysis method of variable length sequence proposed in this paper.The statistical features of the MFCC features of the processed speech sequence are extracted as feature vectors for speech sentiment analysis.(4)The convolutional neural network model is designed,and the speech sentiment analysis method of variable length sequence is proposed to process the speech sequence.The MFCC feature of the processed speech sequence is extracted,and the dimension of the MFCC feature is adjusted to the specified size.For small sample problems,this paper The adjusted MFCC features are enhanced with the characteristics of the speech signal,and then the speech sentiment analysis is performed.Finally,the paper analyzes the experimental results and compares them with the relevant research results,and concludes that the proposed variable length speech signal processing method has certain reference.