Research On Arithmetic Of Speech Recognition Based On Speech Control Vehicle

As the development of speech recognition technology, more and more technicists set the focus on the area of speech control. The key of speech control is that the robot does not only need to get the meaning of the orders, but also can know what to do according to the orders. Based on the speech control technology, small-scale vocabulary speech recognition system is discussed in this paper.Two main aspects are included in this paper. On the one hand, the principle of speech recognition and the speech acoustical model are discussed. On the other hand, an introduction of soccer robot platform and software platform are discussed in the paper. There are three main parts of speech recognition algorithm in this system: endpoint detection, feature extraction, template matching. In the endpoint detection part, the algorithm of short-time energy and short-time zero-crossing rate is described and the result of emulation is given. Based on the method of short-time energy and short-time zero-crossing rate, an improved algorithm is emulated. The paper compares these algorithms according to the emulation data. In the feature extraction part, Linear Predictive Coding,Linear Predictive Cepstrum Coding and Mel-Frequency Cepstrum Coefficients feature parameters are discussed. Comparing the emulation data of these three algorithms, the paper gives the results of comparation. In the template matching part, just the same as the recognition part, the algorithm of Dynamic Time Warping, the improved algorithm of Dynamic Time Warping and the Hidden Markov Mode algorithm are discussed. Because of small-scale vocabulary and isolated-words, the improved algorithm of Dynamic Time Warping is chosen in this system.The programming of the software platform is based on VC-MATLAB, good performance is achieved in experiments.