Research On Speech Interaction System Based On Semantic Enhancement Technology

The speech interaction is a promising human-computer dialogue media,which has rich application scenarios in life and business.In this paper,the main research method is trying to limit the dialogue system to some specific contexts.In the meantime,with the introduction of voice interaction system,trying to determine context of topics and utilize the additional semantic knowledge of speech recognition for further processing,so as to enhance the accuracy of speech recognition results.The topic of this paper is research of speech dialogue interaction system based on semantic enhancement with limited domain knowledge.The main work focuses on following aspects: firstly,analysis of the contextual knowledge in the field of speech interaction,extracting and setting linguistic context and non-linguistic context knowledge.Trying to extract semantic features from language contexts with statistical methods,and using artificial extraction method to set non-linguistic context characteristics,and using the features to identify the context;secondly,analysis of the problems exist in the Pinyin-to-Chinese-Character Conversion model,and proposed using the PLSA to integrate semantic knowledge,which improves the effect of the conversion accuracy from the point of view of statistical language model;thirdly,analysis the rule of additional semantic of interaction information,rules are added to the system to utilize additional semantics in specific areas,and to guide the system to locate and detect errors in order to achieve artificial semantic enhancement.Finally,based on three semantic strategies,the paper proposes the speech interaction system framework with use of semantic information,which improves the speech recognition accuracy and optimize the system performance.The paper designs some experiments for semantic enhancement strategy,and using Pinyin-to-Chinese-Character Conversion accuracy rate,perplexity to analyze the experimental results.The experimental results show that,with the help of the strategy of semantic enhancement,the algorithm can effectively improve the performance of interaction system.