Design And Optimization Of Embedded Speech Recognition System Based On HTK

As voice is an important means of interaction for human beings, voice recognitiontechnology, with rapid development at present has been more and more popular among peopleof all ages. People want to use it to achieve more convenient human-machine interaction andcommunication, to improve human-computer interaction experience. Although speechrecognition technology has been developed more mature than before with a number ofcommonly used algorithms and applications, but the quite real-time and accuracy are limitedin small embedded systems of limited processing power and capacity; therefore, theembedded speech recognition technology research is meaningful and has broad applicationprospects. In this paper, based on embedded platform ARM9, the author takes advantage ofHTK (Hidden Markov Model Toolkit) and has designed a voice recognition system; then,with in-depth study, the author has improved and optimized the system through front-endprocessing in order to make it applicable.The paper includes the following parts:First, this paper demonstrated the design of HTK-based embedded voice recognitionsystem, mainly including setting up the platform and working environment, giving thespecific system implementations.Second, by simplifying the model and using endpoint detecting, the author optimized thetraining and recognition speed, and improved the overall system performance.Third, filter designed for reducing the noise in the speech signal, which improved therecognition rate, furthermore, the application of OOV (Out of Vocabulary) rejection reducingFRR (False Acceptance Rate). Therefore, the improved speech recognition system can beused in practical applications.Fourth, the author used two specific examples to evaluate the performance of this voicerecognitions system: the robot voice command system and human-computer puzzle games,and the results show us the correctness and feasibility of the improved system.