Design Of Robot-Oriented Speech Interaction System

With the maturing of robotics technology,service robots have gradually moved towards homes.As the most convenient and quickest input method,voice has unparalleled superiority.Service robots with voice communication function are more and more popular and loved by users.The development of speech interactive robots suitable for home use has become an inevitable trend in the development of service robots.The difficulty in the development of speech interactive robots lies in speech recognition,especially speech recognition with accent.Therefore,on the basis of establishing a spontaneous speech recognition model,the subject has designed a robot-oriented speech interaction system.The main research work and achievements of the paper are as follows:(1)The basic theory of speech interaction was overviewed.Firstly,the humancomputer interaction technology was overviewed.Secondly,the speech recognition technology was detailedly analyzed,the modeling methods were described about acoustic model and language model.Finally,the speech segmentation technology was told.(2)A scheme was proposed about design of robot-oriented speech interaction system.Firstly,the overall goal of the system was analyzed according to the project development requirements,and the system was analyzed for functional and non-functional requirements.Secondly,based on the requirements analysis,the system outline design was implemented,including the system framework and system flow.Finally,the development and operating environment of the system was researched,and the security of the system was analyzed.(3)A speech interaction system was implemented.Firstly,the spontaneous spoken language speech recognition module was modeled,the acoustic model based on GMMHMM was trained by using the principle of maximum likelihood,including the training of the monophone model and the triphone model,and the statistical language model based on n-gram was modeled by using a large amount of text corpora.Secondly,using the time-domain feature analysis of speech signal and the monosyllabic speech model,the automatic segmentation technology was implemented to solve the problem of filled pause and incoherence recognition in spoken language pronunciation,and using multi-thread decoding technology to improve the efficiency of speech recognition.Thirdly,the design of the client was implemented,including the design of login,registration,recording,and feature extraction modules.Finally,the detailed design of the server was implemented according to the system architecture,including an information management module,a decoder,an online dialogue module,and a file storage module.The results show that the system has a very good recognition effect under mixed noise,and has good robustness,security,and real-time performance.The system provides central module for the intelligent service robot platform.