Research On Voice Control Technology Of Mobile Robot In Noisy Environment

With the rapid development of automation technologies, robots become more and more powerful, and increasingly more products of robots come into our daily life. In such a background, it becomes difficult to achieve the cooperation between men and robots by using traditional ways of man-machine interaction such as mouse, keyboards and buttons. A more convenient and faster interaction way is expected, and the speech recognition technology is able to achieve this. The speech recognition technology is one of the most important branches in the research field of robots, and its goal is to enable robots to understand human languages and provide a more convenient and efficient way of man-machine interaction. The noises in the real environment significantly impact the performance of speech recognition systems. The robustness of the system decides that whether the speech recognition technology can go from the researches in laboratories into the large-scale commercial applications. This thesis analyzes and concludes multiple existing robust speech recognition technologies, based on which we propose a new method of robust speech feature extraction. By using this method, we achieve the speech control for mobile robots under noisy environment.Firstly, we comprehensively study the basic theory of speech signal, based on which we present simulation results of speech signal preprocessing technologies such as pre-emphasis,framing and windowing, and we also introduce the methods of time and frequency analysis for speech signal. Comparing with the characteristics of several common speech feature parameter extraction algorithms, we conduct the Mel-Frequency Cepstral Coefficients (MFCC) extraction based on the speech signal preprocessing.Secondly, we introduce the procedure of speech recognition and deeply study the key technology of speech recognition model training algorithm. Further, we introduce the effects of the speech and acoustic models. We then investigate the Hidden Markov Model (HMM) comparing with the characteristics of several common acoustic model training algorithms.Thirdly, the recognition performance will be degraded in a real speech recognition system due to the mismatching between the environment of speech model training and the practical use environment. Considering this fact, we study the robust speech recognition technologies and the normalization approach of feature parameter. We then propose a robust feature extraction method that combines the feature parameter normalization and time series filter. Based on the MFCC feature parameter extraction, the robust speech feature extraction is conducted and validated by experiments. Experiment results show that the proposed method is able to significantly improve the accuracy rate of the speech recognition system under a noisy environment and enhance the system robustness.Finally, we study the construction approach of the mobile robot that is based on the Arduino Mega 2560 single-chip microcomputer, and we investigate the motion control and Bluetooth communications for the robot. Based on the proposed robust speech feature extraction method, we use the left-to-right HMM with 5 states as the acoustic model and develop an Android based speech recognition software. By using this software, we implement the speech control for the mobile robot under noisy environment.