Research On Speech Recognition Of Mengjin Dialect Based On HTK

Speech recognition technology to the intelligent era of people brought great convenience,enhance the efficiency of human-computer interaction greatly.Especially in today's mobile Internet era,intelligent mobile terminals and other intelligent devices have a wide range of popular,for speech recognition technology research needs larger than before.In Chinese speech recognition,Mandarin and major dialect language recognition is already quite mature,but in the real life of the use of scenes,even if the same language dialect will be very different,which will affect the correctness of speech recognition and Robustness of the system.How to improve the adaptability of the system to the accent of a variety of complex environments and the speaker's own phonemes is still a problem.And now for some dialects of speech recognition research has not yet deep into such a small language branch.Based on the above reasons,as well as the author's home in Mengjin,is very familiar to Mengjin dialect,easier access to a large number of voice data,so selected this topic as a research topic.In this paper,the speech signal preprocessing,feature parameter extraction and so on were studied,and built based on HTK Mengjin dialect speech recognition system,the main contents are as follows:(1)Elaborated the background and significance of the topic,introduced the development of speech recognition technology and the status quo.(2)Overview the speech recognition system,Introduce the related technology and principle of speech recognition system,from the establishment of corpus to the preprocessing of speech signal and the principle of feature parameter extraction.(3)Introduce the model of speech recognition,focuses on the idea and definition of hidden Markov model,and gives the solution of three basic problems of hidden Markov model.(4)Introduce the HTK toolkit and its working principle,and gives the compilation process of HTK toolkit in Windows 7 environment.(5)This paper introduces the process of constructing the speech recognition system based on HTK,gives an example to verify the reliability of the system,and find 10 local people with different age,gender and educational level to record 200 training voice data and 20 Test voice data,set up based on HTK Mengjin dialect speech recognition system,the experimental simulation,The word recognition rate is 80.95%under the condition of 39-D MFCC characteristic parameter extraction,77.78%is obtained under the condition of 13-D MFCC characteristic parameter extraction,and then the comparison is made under the condition of Wechat recording in the experiment,the recognition rate of words is 79.37%under the condition of 39-D MFCC dimensional characteristic parameter extraction,and 68.25%is obtained under the condition of 13-D MFCC characteristic parameter extraction.