The Research Of Emotion Recognition Integrating Text And Speech Features Based On Deep Learning

The researchs on the fileld of human-computer interaction have been an important aspect of intelengent technology application research. Natural language understanding is the significant way of completing in human-computer insteraction and the key technology of emotion recognition for implementing natural human-computer interaction. As many interdiscipline and computer science having corresponding development, the research mission on emotion recognition task which is an important field of artificial intelligence in theoretical research ant practical application which has been great developed.During human-computer interaction, the task of the facial emotion expressions recognition mainly faces the problem: signal acquisition and information processing. The method has higer demand to device. The requisition of speech and dialogue content belong to the same path, text information belongs to voice carrying content, these two processing has the same identity thus the speech signal emotion recognition method and the signal processing methods can be classified based on the text. Under the background of the study and considering the individual treatment and the limitations of language and text, the paper will combine two kinds of language and text mode for emotion recognition. The main research contents and work are as follows:1. First is the collation work of speech emotion database, the paper sets mobile customer conversations as the experience corpus and tags emotion label of data, and then turns these speech data into text. Next step is that design a mehod for sentence segmentation of whole speech conversation.2. Research on emotion recognition based on text signals. We use several statistical methods and semantic layer features to accomplish the task of text feature extracting. According to the characteristics of the essay emotional expression and combing with different forms of text, the paper mainly expressed in the extraction of the text as far as possible from the emotional information for completeness.3. The research of speech emotion recognition, in this paper, two kinds of different characteristics of speech signals are extracted based on metrics (based on emotional characteristics and emotional feature extraction).4. Based on deep learning modal characteristics and structures we build depth environment to fusion two kinds of model characteristics and execute model training and use various fusion methods for integratation of text and speech features After that we comparison experiment, verify the feasibility of the method in this paper.