The Research Of Speech Recognition Based On Algebra Algorithm And HMM

In today's information age, with the rapid development of computers, all kinds of intelligent machines have also entered all aspects of our life, causing people eager to communicate with the machines can be as natural as the fluid exchange between people. Due to the practical value speech recognition has received more and more attention recently. Speech recognition has a broad social and economic benefits, and has achieved some efforts, but it still has many problems in the process of practicality. People also need deeply research the theory and algorithms of the speech recognition to make practical useing of them.In this paper, for increasing the rate of the speech recognition, it has provided a comprehensive study of theoretical basis for speech recognition. And it introduces a novel learning algorithm——algebra neural network. From experiment result, the algebra algorithm compared with the backpropagation algorithm is much better.HMM has the strong time domain modelling ability, but the classification is weak; while the algebra neural network is just the opposite, it has strong classified ability. In order to overcome their faults, algebra neural network and HMM is combined in this paper. The algebra neural network instead of the gauss mixer to calculate the observation probability. Finally, it design a small-vocabulary continuous speech recognition system based on the mixed model, through experimental verify the validity and serviceability of the mixed model.The in-depth study and practice have been conducted in this paper, especially, the exploring works about recognition methods have been made, and have made certain progress, which will laid a foundation for the speech recognition research.