Research On Speech Separation Technology Of Mobile Robot Hearing For Navigation System

With the continuous development and improvement of digital signal processing technology and acoustic electric technology, as well as the level of humanoid robot improving in artificial intelligence, auditory system, as an important component of the human senses, has become an important study in robot research. Since the voice has the characteristics to bypass the barrier, in robot multi-information acquisition system, hearing can be matched with the robot vision to make up its limited field of vision and the limitations that can not pass through the non-translucent barrier. In complicated task models, high-quality voice is a prerequisite that the human-computer interaction can rely on hearing system of intelligent robot. Therefore, the acoustical perception system based on speech separation technology can widely apply to actual noise circumstance and supply clear and independent source signals to intelligent robot for its hearing system as keen as humanoid audition.The traditional signal convolution model of blind source separation algorithm is improved through the use of binaural model based underdetermined blind source separation algorithm in this paper to solve the real-time problem produced by iteration for separation matrix, further enhancing the adaptability of blind source separation algorithm in actual noisy environment. In order to increase the searching performance of robot hearing system in three-dimensional space a sound source localization and navigation system of mobile robot hearing based on time delay estimation and underdetermined speech separation is presented in this paper. The whole system should be divided into four parts including the signal acquisition and preprocessing, speech separation, position calculation and robot control. The tetrahedral structure of microphone array can percept signals from three-dimensional space. The system can estimate source signals using a real-time underdetermined blind separation method based on time-frequency masking which constructed by the histogram of parameter space. The measurement of target position depends on mathematical model of position calculation. Under the dual control of programs, find the target and avoid obstacles, robot can avoid obstacles and reach the target position successfully. Experiments showed that the design can make the robot find the position of many sources and reach the target rapidly and accurately.