Research Of Algorithm In Identifying The Speech Recognition Based On Neural Network And HMM

With the rapid development of computer technology, the computer has entered every household to bring endless human convenient. At the same time, the advent of the information age has led to increasingly high demands intelligent computer, which is also reflected in computer interfaces. Interactive allowing the "machine" understand the language and are concerned that speech recognition technology is the realization of this function, in reality many industries have adopted speech recognition technology intelligent management, the speech recognition and speech synthesis of modern computer technology has also become an important research and development areas.Speech recognition technology and multidisciplinary research areas have links to those areas of scientific research achievements have become an important factor in the development of speech recognition technology. Speech recognition technology has made some achievements, but because speech signals diversity and complexity of current speech recognition efficiency is not high. Efficient identification and development of models and algorithms speech recognition research into an important topic.The first speech signal processing and feature extraction issues discussed from thethree parameters-effective speech recognition features LPC coefficient, thenumber of LPC do discovery and Mel frequency opposing discovery requirements (MFCC);Secondly focused on speech recognition, the three commonly used methodsof identification------ based on template matching of the vector quantified (VQ), theneural network models and hidden techniques in the application of speech recognition, respectively, discussed their training and model identification algorithm for Hmm algorithm detailed discussion;Then combined neural network and Hmm make improvements based on statistical methods that the neural network models and algorithms /HMM of speech recognition;Finally , through simulation experiments comparing parameters, algorithms for the identification of the different effects and the isolation of speech recognition for speech in the discussion of the effects. MFCC parameters than as a feature that LPCC as to the high rate of identification of parameters in isolation speech recognition, the use of VQ model algorithm than use algorithms to identify PNN/HMM much higher rate, but in continuous speech recognition, the latter figure to the high recognition rates.