Studies On Robot's ASR Based On Embedded And Cloud Technology

With the rapid development of robot technology,traditional mouse and keyboard in the old days and other man-machine interaction ways cannot continue to meet people's demands any longer,so seek a brand new man-machine interaction way is the goal every researcher claims to be pursuing.Along with the continuous improvement of speech recognition and the development of cloud computing over the years,it makes the increase of new man-machine interaction ways possible.However,robot's embedded devices are confronted with challenges in computing speed and storage capacity:mass data and speedy computation cannot be achieved.What's more,the recognition result also leaves much to be desired.Therefore,the study on the embedded robot's ASR has practical significance and broad application prospects.The main work of this paper is based on ARM9 embedded core board,designing and optimizing its periphery circuit.By using pocketsphinx ASR toolkit and Baidu cloud ASR technology,design an embedded ASR system with network and one without network respectively,and both these two systems can realize ASR functions.Hence,the main research contents of this paper includes as follows:(1)Systematically introduce this research's background and significance,ASR research status at home and abroad as well as the challenges that robot's ASR faces.(2)According to the generation of speech signal,describe the recognition process of ASR technology in detail,and then elaborate speech signal's preprocessing procedure and its feature extraction method.(3)Introduce the embedded platform used in this research,and illustrate the used hardware resources used in this platform and embedded platform's building process.Finally complete the burning of system.(4)Introduce PocketSphinx ASR toolkit in detail,and achieve the building of PocketSphinx ASR training platform on the PC end.Complete the training of both acoustic and language model,and finally finish embedded transplantation.(5)Propose an embedded cloud ASR solution based on the defects of embedded devices,and it mainly include Baidu cloud ASR API's building and embedded transplantation.(6)Summarize the embedded ASR solution adopted in this paper,and complete the continuous speech recognition work with large vocabulary and small vocabulary,which has very good practical performance.