| Recently,speech recognition technology has been rapidly developed and widely applied in many industry practices, such as automation,aerospace,information Communication, etc.. General algorithms of speech recognition include Dynamic Time Warping, Hidden Markov and artificial neural network. Those algorithms mostly take long time to process recongnition and calculation for large amount of data, which causes more difficulties for industrial applicaiton. Around this problem, we propose a speech recognition algorithm based on vowel period to realize automation geometric model. This algorithm completes the template matching according to the characters of vowel sequence, and can significantly reduce the amount of calculation and the recognition time when the database is relatively fixed. However, the accuracy of this methord is much lower than the currant algorithms(the accuracy can reach 95%). Then in the practical application, this thesis still uses the hidden Markov algorithm for the automatic speech geometry modeling. This thesis provides active exploration for industrial design interactive approach. Therefore, this thesis has important application value.The application target of this thesis is auto modeling by use 3D modeling software. It uses Hidden Markov algorithm as a representative for speech recognition study. Around the design automation, this thesis mainly includes:Firstly, this thesis studies the Dynamic Time Warping algorithm and the Hidden Markov algorithm for speech recognition. The thesis includes the fundamentals and relevant technologies of speech recognition for the two algorithms and we build a speech recognition system. In that system, the thesis uses that two algorithms to simulate experiments and compared the results. The results show that Hidden Markov time algorithm is more efficient than Dynamic Time Warping algorithm. The thesis also does tests for a speech recognition algorithm based on vowel period. The results show it is hard to distinguish the words just by a vowel sound signal probability distribution. At the same time that result in the very low accuracy when using that algorithm in speech recognition. So this thesis uses the Hidden Markov algorithm for speech recognition of three-dimensional modeling.Secondly, this thesis studies the classification and identification of feature elements in three-dimensional modeling software. The modeling samples are classified by the length of the words and the test template is classifed into the corresponding length of samples for speech recognition. However, the tests result show that this kind of recognition method will reduce the overall accuracy. So the thesis improves the word samples, then the accuracy of the speech recognition is improved a lot. In this thesis, we select the menu item coordinates in one three-dimensional modeling software interface as the geometric feature. By exchanging data between tables, the result of the speech recognition can match the geometric feature and complete the speech control for that three-dimensional modeling software.Lastly, this thesis builds a prototype system for automatic geometric modeling through voice control. The main task of this prototype is to complete automatic geometric modeling by voice control in the Pro/E software. This thesis completes recording software development, dynamic link library call for Speech Recognition, and call for the Pro/E software, respectively. Then we complete the prototype system module integration development. After that, the experimental analysis and performance testing of the prototype show that the prototype system can realize to control the Pro/E software by voice basicly. This prototype system takes 20-37 seconds to finish one test. |
PDF Full Text Request