| Compared with the traditional key operation and touch screen operation,the on-board infotainment system based on voice control reduces the driver’s line of sight shifting and the number of hand-to-hand shifts,improves traffic safety,enhances the user experience,and gradually becomes the standard and selling point of mainstream models.However,at present,voice control still has problems such as relatively low accuracy of speech recognition,relatively inflexible voice commands,and poor user experience.Therefore,it is of great theoretical and practical value to study spoken language-based speech control systems.This dissertation proposes a semantic understanding framework based on a combination of template and statistics.The framework supports both semantic understanding based on rules and semantic understanding based on statistics,making the two methods complementary to each other.Realize human-computer interaction based on natural language.The main work and innovations are as follows:First,the design and implementation of a semantic understanding module based on semantic templates.Defining semantic templates based on the functional requirements of the onboard infotainment system and the interface of each functional module;collecting different sentence patterns based on the functions to be implemented;arranging sentence patterns into regular expressions and storing them in XML format to implement regular expression based Matching technology template matching semantic understanding algorithm;design based on the Tire tree dictionary and keyword-based index-based fast lookup algorithm to enhance semantic understanding performance.Second,the design and implementation of named entity recognition based on rules,dictionaries and statistics.Explain the basic principles of rule-based naming recognition,implement rules-based fuzzy identification of contact names,elaborate the basic principles of dictionary-based named entity recognition,implement dictionary-based accurate identification of artists and songs,and expound the basic principles of statistically-based named entity recognition,Marking place names and broadcasting station name and word materials,realizing geographical name recognition based on conditional random fields and radio station noun identification.Third,the design and implementation of a statistical semantic understanding module.Combining the results of Chinese word segmentation and named entity recognition,feature extraction of key words and key semantic classes based on Chi-square test algorithm and feature weight calculation based on TF-IDF are realized;and based on this,text classification based on support vector machine is realized.Marking information extracts corpora and implements extraction of semantic parameters based on conditional random fields;finally achieves semantic understanding based on statistics.Based on baidu speech recognition engine,mandarin is used for testing in a quiet environment,the experimental results show that the accuracy of semantic understanding of navigation,telephone,radio and music functions reached 76.83%.There is a certain theoretical significance and application value for the natural language understanding facing the car. |