Back-end Processing Of Automatic Speech Recognition By Semantic Activity Algorithm

In recent years,during Speech Signal Processing,the application research on Automatic Speech Recognition(ASR)technology becomes more advanced,and the Statistical Method of Natural Language Processing(NLP)for Speech Information Processing has been improved,while the Rule Method of Natural Language Processing remains weak,with the obstacle of Semantic Understanding encountered on ASR,so that the progress in Human Computer Interaction(HCI)of Artificial Intelligence(AI)has been slow.The main purpose of this paper is to do a reserach on the back-end of ASR by the method of Language Rules to solve the technical problems for fuzzy Sounds such as Chinese homophones in speech recognition,and to complete the deep processing technology of Semantic Analysis for ASR.This paper firstly,outlines both the research background and techniques in ASR and Semantic Analysis,and points out the difficulties and problems currently faced by ASR and Semantic Analysis,illustrating the importance of the urgent need to break through the bottleneck of Chinese Automatic Speech Recognition technology,and presents a technical solution of semantic analysis.Secondly,summarizes the Semantic Activity Algorithm(SAA)based on Chinese Activity Rules.Thirdly establishes a semantic framework and recognition mode for ASR,to deal suitably with Automatic Speech Recognition Fuzzy(ASRF)which has failed to be recognized after processing with the means of appropriate mathematical statistics.Finally,the paper puts forward a semantic analysis method for back-end of ASR for deep processing fuzzy sounds,which consists of activity mark,segmentation,pattern recognition,semantic annotation and other components,and by which the fuzzy sounds are properly processed that are produced from the aspects of sound structure,accent habits,pronunciation errors,spoken umlaut,mixed speech languages,and a beneficial effect has achieved.The Activity techniques of ASR and method of this study have opened up a new pathway of the Rule Technology of ASR on the basis of the front-end processing with Hidden Markov Model(HMM),Bayesian Analysis(BA),Artificial Neural Networks(ANN),Machine Learning(ML)and other statistical technologies of ASR.The innovation point of the paper lies in the settlement,and we firstly put forward an approach for ASR to process difficult Fuzzy Sounds including homophones and similar sounds as a frontier technology of Semantic Analysis for ASR which lays a solid foundation for HCI in Language Understanding of ASR with Rule Method of Natural Language Processing.The paper also points out the relevant issues that require a further study on ASR such as the semantic association of Speech Context.