Study Of Speaker-independent Speech Recognition And Robot Voice Control Based On ANN

Speech recognition is to make computers understand semantic content of humanlanguage. It is a comprehensive discipline technology, which contains acoustics, linguistic,signal processing, computer technology and artificial intelligence etc. People are eager toachieve effective and efficient human-computer interaction in man-machine interface withthe wide application of speech recognition and artificial intelligence, the study ofcommunication and interaction between human and robot is becoming a research focus.The research object is for speaker-independent isolated-word recognition (SIIWR), thetwo key recognition technologies will discussed in this paper, which are Hidden MarkovModel (HMM) and Artificial Neural Network (ANN) technology. Combing HMM modelwith ANN model, a new model called HMM/ANN was proposed, the hybrid model takesfull advantage of the strong ability of HMM in time domain modeling and the ability ofANN in self-learning and pattern recognizing domain. The deficiency of the original modelwas improved, and the structure and design process of the hybrid model was described indetail.Then, the hybrid model was programmed and the simulation experiment was done byMATLAB tool, and comparing the other widely used recognition models. Theexperimental results showed the hybrid has improved the speech recognition rate andrecognition performance, and the average recognition rate is up to95.7%. In order to makethe theory practical, robot voice control system was established based on the hybrid model,including one set of simulation system and one set of real-time human-computerInteraction system. The capacity and effect of practical application was validated, theexperimental results showed the hybrid model used in robot voice control system couldmeet the general requirements.