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

The speech recognition is a technology of making machine understanding humanlanguage. Especially in the current field of artificial intelligence, speech recognitiontechnology has become a key technology of man and machine dialogue. Therefore,speech recognition technology has become the focus of the countries in the speech field.At present, speech recognition technology has been widely used in voice dialing, voicecontrol, voice navigation, speech document retrieval, and other fields. So it was got bythe unanimous recognition of the community.In this case, the paper first analyses the process of the speech signal preprocessingand feature extraction, and then emphatically introduce the two speech featureextraction method, linear predictive cepstral coefficients (LPCC) and Mel-frequencycepstral coefficients (MFCC), when doing reseach of speech recognition based onhidden Markov model. After analyze the speech signal pre-processing and featureextraction, gradually get the characteristic parameters that can represent the nature ofthe speech signal, and reduce environmental noise and pronunciation duration andintensity of different factors that may cause the speech recognition errors. And then, thespeech recognition based on hidden Markov model (HMM), the result of recognition bythe HMM algorithm is as a candidate word. Finally, adopt BP neural network to detectthe candidate word whether it is correct. Also, it is to determine whether the candidateword belongs to the corresponding classification. The experiments prove, the methodbased on a combination of isolated word speech recognition by HMM model and the DPneural network can reduce noise and other factors, which cause the HMM model ofrecognition errors, and enhance the robustness of speech recognition systems.