Design And Implementation Of Speech Content Emotional Tendencies Classification System Based On Triple-Play

With the progressing of the Triple Play Project in China，network video, audio andother services will get faster and better development, it greatly enrich the cultural life ofthe majority of consumers, but also with network information security challenges. In orderto be assured that the video, images, text, voice s content on the network comply with therequirements of the national laws and regulations. Triple play content regulation system isrequired. Content regulation system to ensure the realization of the content of the wholenetwork can be controlled.To achieve a safe and reliable triple play of the audio-videocontent playback. Based on these requirements,we proposed emotional tendencies analysisof the speech content,for the timely detection of contraband or suspicious content in theaudio data.Emotional tendencies analysis of the speech content relates to the field of speechrecognition,text classification,text emotional tendencies analysis and so on. We must firstsolve the problem of speech-to-text,the non-specific human large vocabulary continuousspeech recognition technology is one of the key technologies, feature extraction, acousticmodel, language model is the work of the three core. The text emotional tendenciesanalysis aims to determine the attitude or stance of the text content on the topics,completetext topic classification based on text emotional tendencies analysis.text topics and text thetendentious two information to determine whether the text content are prohibited.Use of existing technology to build audio content tendentious analysis prototypesystem,The system is mainly divided into two modules for speech recognition and textanalysis. The speech recognition is implemented by the University of CambridgeHTK,HTK provides for feature extraction, acoustic model creation, the voice modelcreation tools or interfaces,we can complete the work of the speech-to-text; After study avariety of text emotional tendencies analysis techniques,we have proposed for the system based on the phrases-collocations method,combined the tendentious dictionary and textclassification techniques,We can identify text content expressed attitude " positive" or"negative".The system is still in the experimental stage,put into application in the future it willprovide good support for triple play content supervision.