Study On Angry Emotion Detection In Continuous Dialogue Speech

The goal of this thesis is detecting angry emotion in the continuous dialogue speech,which is to determine time period information of speaker's angry emotion in continuous speech.With the widespread use of emotion calculation and speech recognition in multiple fields,the speech emotional research has gained increasing interest,and angry emotion detection of the speech has a certain practical application needs.So,the thesis achieves angry emotional speech detection algorithm,and designs a detection platform to show the detection in a webpage.The followings are main works:1.The establishment of the angry emotional speech database: base on two Mandarin emotional databases and the THU_EV₁062 database of Tsinghua University,we select angry emotional speech and part of non-angry emotional speech to listen and analysis.And two sub-libraries,contrast sub-library and angry emotional grading sub-library,are established.2.The accomplishment of the angry emotion recognition model: we select and extract some speech features,which can characterize the angry emotion better,such as MFCC,PLP-CC,pitch,formant,etc.And the final angry emotion recognition model is trained by soft-voted aggregation of two classifiers,the SVM classifier and random forest classifier.3.The detection of angry emotional speech in continuous dialogue: after the pre-training of the input continuous dialogue speech,the detection of angry emotional speech is achieved by sliding window.And the final time period information of angry emotional speech is extraction through maximum filtering and emotion state limited filtering.4.The grading detection of angry emotion: according to the emotion level in psychology,the angry emotion is divided into two levels.And the angry emotional grading model and grading detection subsystem are established.5.The establishment of the speech detection platform: with the web mode and the computing support by server,we establish speech detection platform.We build the web framework of the platform by Flask modules of Python with JavaScript to achieve the front web page preparation and HTTPS protocol to realize the transmission between server and browser.The embodiment detection is completed on this detection platform.The test result verifies that the algorithm completed in the paper is effective.