In order to meet the diversity requirements of mobile robot interaction control and improve the speech control performance of mobile robot, a speechbased mobile robot control system is designed. Through analyzing the control signal transmission process and speech signal noise source of the robot in using environment, the composition scheme of the mobile robot speech control system is made up. The main flow for the implementation of front end speech recognition section is given. And the dereverberation algorithm for speech enhancement is emphatically designed. By fully utilizing the potential spectral features of speech, a dereverberation algorithm is proposed based on combined nonnegative matrix factorization and deep neural network. Firstly, the speech signal features are obtained through matrix decomposition, and then the feature vector is generated to train the activation function, which reduces the training complexity of the deep neural network model. The comparison analysis shows that the proposed algorithm possesses superiority in solving the speech reverberation problem. The control software was written, which was embedded in the speech recognition algorithm. A speech control platform of industrial mobile robot was built to verify the effectiveness of the speech control system. In the reverberation environment, speech control experiments were conducted, in which different people performed multiple actions on the robot. The results show that the system can realize the speech control of mobile robots. The proposed speech recognition method can achieve the average correct execution rates of actions of 96%, 95% and 93%, respectively for the mobile robot under the reverberation conditions of 03, 06 and 09 s.