| Nowadays, as the speech recognition gain the more and more wide application,the key words recognition because of its identification method and application field ofparticularity, also become the focus of attention in recent years. during theimplementation process of prison sentences, according to the prison regulations,the guard monitoring the serve personnel voice during the meeting between thepersons serving sentences and its relatives. Because of all kinds of dialect from thepersons serving sentences and its relatives, the prison police have larger thelanguage barrier in monitoring, also add a lot of police and time.This paper puts amethod forward the key words of the dialect speech recognition can automaticallyidentify key words in the dialect. the guard can to understand the general statementmean through the keywords, Master the content between criminal and itsrelatives,and reduce the workload of guard greatly, which is to overcome thedifficulties brought by the lack of understanding of the dialect more importantly.Facilitate the prison police target to carry out education modification work from thecriminal psychology and family status, solve various contradictions, make it toachieve a new life as soon as possible.In the realization method, the method of key words recognition is similar to otherspeech recognition methods, which have DTW, HMM, ANN and so on, themainstream method is hidden markov model (HMM), in this system, the HMMmodeling method for modeling and identification of key words is adopted. Theresearch content mainly includes:1. In the speech endpoint detection, We use a method of combining the energyand short-time zero-crossing rate to segment a long speech;2.We adopt the Viterbi search algorithm, and figure out the maximum a posteriori probability in the keywords model through the HMM identificationalgorithm;3.We constructe a key word recognition system. in HMM speech recognitionmethod based on the theory,We use the Visual studio2008platform,and design anddevelopment for a speech recognition system about the specific person andkeywords through the console. |