| Chinese speech recognition is a very complex task. To remove the complexity ofspeech recognition technology, the complexity of the Chinese dialects also creategreat difficulties for the promotion and application of speech recognition. At present,researchers not only study Chinese Mandarin recognition of the voice, but also somepeople began to voice recognition of the local dialect. Dialect speech recognition inthe country has just begun, immature technology, many of the key problems to beresolved.In recent years, hidden Markov Model (Hidder Markov Model, HMM) in thesuccessful application of the speech signal processing greatly promoted thedevelopment of a speech recognition technology. Automatic speech recognitiongradually from the research direction of the specific people, small vocabulary,isolated words to speaker-independent speech recognition, large vocabulary,continuous speech recognition transfer direction.In this paper, the hidden Markov model-based the Yongzhou dialect digitalidentification system, complete the following main tasks.(1) Systems use Yongzhou Lingling voice standard dialect speech. Select thethe Lingling people pronounce as a standard in recording, analyzing characteristics ofthe the Yongzhou dialect pronunciation, and the pronunciation marked out theirdifferences, to extract the speech features. Create a small corpus of the Yongzhoudialect digital speech recognition.(2) Established a Yongzhou dialect digital voice acoustic hidden Markov model,and model optimization.(3) Respectively using a different primitive, different feature dimensions on the"0-9"10the number of training speech and test speech recognition experiments.Experimental results and analysis, found that26-dimensional one syllable recognitionrate, internal and external recognition rate reached92.61%and91%in SpeechRecognition, which is a good experimental results. Experiments show that theestablishment of the the Yongzhou dialect speech recognition system is feasible. |
PDF Full Text Request