Research Of Speech Recognition Based On The Combination Of Hidden Markov Model And Artificial Neural Network

In the development of the computer technology and information technology era, all kinds of intelligent machines also gradually in people’s life plays a important role, people hope and natural communication and exchange, this also makes the speech recognition has make a spurt of progress of development. Speech recognition is a promising, social benefits, economic benefits obvious technology widely, in the rapid development at the same time, but also because of the experiences from the laboratory to the process of practical application, therefore inevitably have some problems.Based on the advantages and disadvantages of HMM in hand, analyzed from the modeling and decision-making abilities of different perspectives, that of both advantages and disadvantages. In order to get the artificial neural network algorithm is used to solve HMM inferior way. Artificial neural network can solve the classification ability of HMM is weak, the analysis describes a hidden Markov (HMM) model and radial basis function neural network (RBF) after the fusion effect in order to realize speech recognition. On this basis, in order to show that the hybrid model of the algorithm is correct, based on the analysis of the calculation method of mixed model simulation model to the MATLAB environment, according to parameters, not the same number of training samples, the background noise in terms of this model is really able to speech recognition. Then, using CDHMM model and HMMNN model using isolated Chinese digital speech recognition experiments in a wide range of people, to prove which model is relatively good practicality. From the two experiments we can find that, the model proposed in this paper to achieve the function of speech recognition is correct, but the HMMNN mixed model is more effective than CDHMM speech recognition, enhance the robustness of speech recognition system.