Realization Of Voice Interaction And Semantic Recognition For Robot

Natural language interaction is one of the hotspots in the field of artificial intelligence research.More and more voice interactive platforms have gradually entered our lives.For example,there are plenty of intelligent voice assistants developed by foreign countries,like Apple Siri,Microsoft's Cortana,Google's Google Now and so on.Domestic enterprises such as Xunfei,Baidu and Tencent are also developing their own voice interactive entry.Natural language interaction includes speech recognition and semantic analysis.This paper is also based on these two aspects..In the aspect of speech recognition,we introduce the general framework of speech recognition,analyze the theory of Hidden Markov Model,and then elaborate the detailed process of speech recognition based on hidden Markov model using HTK,giving the experiment and analysis.In semantic analysis,we introduce the related theory of natural language processing,and use convolution neural network for semantic analysis.The traditional convolution neural network is composed of consecutively linear convolution layers and pool layers,which can extract features from images,speeches and texts.We propose to replace the traditional linear convolution filter with the cascade convolution layer to improve the expression ability of convolution neural network.In this paper,we take emotional analysis and problem classification as an example to use the improved convolution neural network on semantic analyzing.Experiments show that the improved convolution neural network is better than the traditional convolution neural network in the case of training speed as well as of recognition rate.In addition,we also design a robot based on ARM interactive voice response system.Considering poor computing ability and small memory space of the embedded system,the voice recognition module and the semantic analysis module are trained in the personal computer and then transplanted to the ARM development board.