The Research On Speech Emotion Recognition Algorithm Based On BP Neural Network

Speech is important means in communication between people and it is one of the fundamental methods of conveying emotion,on a par with facial expression.Speech signals covey semantic information,meanwhile,they also transmit emotional information,moreover,emotion plays an important role in communication.So along with rapid development of Human Computer Interaction system,emotionin speech is a topic that has received much attention during the last few years,in the context of speech synthesis as well as in automatic speech recognition.Emotion has played a significant role in the process of human decision-making and perception.For a long time research on emotion intelligence has only been done in the fields of psychology and cognitive science,but along with the rapid development of information technology and the growing concern of relationship between human and computer these years,how to achieve personification of computer,which can apperceive our environment,our emtion etc,which has become the important sign and goal of man-machine interactive ability.The combination of emotion intelligence and computer technology brings the novel research area named emotion recognition. Speech emotion recognition is a key part of affective computing.The emotion features are extracted precisely from the wave signals by computer and used to recognize the emotion state.The paper has conducted the research focusing on speech emotion recognition based on BP neural network.The dissertation is organized as follows:(1) In front of the speech signal processing.Emotion sentence is effectively pre-emphasised,windowed and endpoint detected.Studied the short-term zero-crossing rate and short-term energy extraction method,compared and analyzed the estimation algorithm of Pitch,improved the estimation algorithm of Pitch by studying the methods proposed by previous.(2) Analysis and extraction of characteristic parameters of emotional voice.Analyzed statistically change discipline of characteristics for 120 emotion statement,studied characteristic information about emotion,certained 16 eigenvalues for speech emotion recognition,formed a 16-dimensional feature vector,including: the maximum,minimum value,medium value of the first formant,the second formant,the third formant;the maxmum value of Short-time average zero-crossing rate;the medium value,maximum,minimum value of pitch frequency and the maximum,minimum value,medium value of short-time energy.(3) Since the extracted feature vector is high dimensional,and has some relevance,that is,there is a certain degree of redundancy.Therefore,this article focusing on the neural network training samples were normalized,and then made a principal component analysis,not only reduced the input of the feature vector dimension,but also removed information in addition to.The paper analyzed the structure,principle and shortcoming of BP neural network.In experimental environment with MATLAB6.5,two improved BP algorithm is used to identify emotional voice,with the traditional BP algorithm,two improved BP algorithm improved the recognition rate and convergence speed.