Embedded Human-machine Speech Interactive System Based On Cloud Service

With the development of embedded system and information technology,the application of speech interactive technology on embedded system has become the hotspot of research.Speech interactive technology mainly includes speech interactive equipment,speech recognition,speech synthesis,speaker identification,and so on.Furthermore,the developing of deep learning and cloud computing promotes the speech interactive technology effectively.So it's valuable to research and implement the design of cloud service based embedded human-machine speech interactive system.After the introduction of fundamental theory of speech signal processing,this thesis focused on the source localization technology based on microphone array.A cross-square planar orientation quaternion model is designed,the source-locating equations are deduced using the estimated delay information.Low pass filtering and clipping technique are introduced in the generalized cross correlation algorithm to improve the estimation precision of speech signal delay.Furthermore,linear prediction is used to whiting the speech signal before the generalized cross correlation algorithm is used to estimate the delay information,which can further improve the estimation precision.An embedded human-machine speech interactive platform based on TMS320DM368 DSP is designed using modularity design technology,which includes main processing module,power management module,data storage module,network communication module,speech acquisition and playback module,and so on.After that,the transplant of embedded Linux operating system and the design of related device driver are realized.Application software is developed on the constructed platform,which implements multi-channel speech signal acquisition,and sound source localization,speech recognition and speech synthesis based on cloud service by calling the Microsoft Azure interface,speech playback,and so on.Performance testing for the designed embedded speech interactive system is executed under laboratory environment.The testing results show that the designed source-locating algorithm can provide the estimation precision of azimuth and elevation with 10 degree,the commands can be recognized correctly,and the speech synthesizing and speech playback are implemented normally.A framework of intelligent human-machine interactive system is constructed,which provides some foundation for subsequent research.