| Language Identification (LID) is the technology to identify the language by usingcomputer to automatically cognize and judge speakers’ speech sound. With the developmentand widely application of the speech recognition technology, the language identification withpolice ends is of great use and with prosperous future in the front-end multilingualrecognition, the early warning of intelligence, collecting information and fighting againstcriminal crime. Yunnan province is inhabited by a few ethnic groups. It is a fact that it isdifficult to communicate between ethnic groups because of language specifics and the lack ofexperts in ethnic group languages. This makes it an important and urgent task to conduct theresearch and develop a language identification system that can meet the needs of policeoperations connected with the recognition of ethnic group languages in Yunnan.The system uses pattern matching as the method of speech recognition, and makes fulluse of the technology of Vector Quantization(VQ) and Dynamic Time Warping(DTW), toachieve the langugae recognition and key words recognition of different minority langugagesin Yunnan. Firstly, the system employs Mel Frequency Cepstrum Coefficient (MFCC) as thecharacteristic parameter of linguistic identification, with the aid of VQ technology MFCCwould become the codebook of dialectical speech features, and calculates the averagedifference between the codebook with the MFCC of a speech in issue in order to find theminimal quantitative error and pin down the most correspondent linguistic type down. By Secondly, the system selects the keywords that match our requirements to produce StableFrame Veictor Matrix(SFVM), and calculates the speech that are ready for identification so asto getting Test Frame Veictor Matrix(TFVM). Next, The distance between SFVM withTFVM is figured out by DTW in order to get Distance Matrix(DM) and the similarity with thefixed keywords, so that it could identify the keywords of a speaker. Lastly, according to theexperimental data, it employs different methods to revise language codebook and keywordssample, which promotes the efficiency of systematic identification. The experiments showthat MFCC is good at depicting human being’s auditory features, and VQ is useful forcollecting the parameters of the speech of different genders, and DTW gains betteridentification results, given the types of features are properly chosen.The development of the system takes about a year. After the16local public securityoffices including Kunming finished the installment, disposition and debugging of the relatedhardware and software one after another, the Yunnan Police Minorities LanguageIdentification System achieved concentration of the construction at the province level. Atpresent, the system has come into service online for police operation in Yunnan. By the secureaccess and interaction with other existing police information systems, the languageidentification system has completed the efficient access and query to the speech database. Thesystem has shown a good capability in language identification and key word recognition tothe ethnic group languages which have urgent needs to be identified at police operations. Itpromptly indicates the direction for detective work in related cases, and even possible to lockthe criminal suspect. The system also plays a good role in collecting information, analyzingintelligence and giving early warnings for security work for some important event withspecific time and location.This study is the first in public security field to deal with language identification forminority languages in Yunnan. As there are no existing systems and results to compare with,the system under the study might be insufficient in selecting speech samples and extractinglanguage features. Besides, the realization algorithm still has the potential to be improved.We believe that the system can be further improved relying on the specific needs at work andthe experience with the system. |
PDF Full Text Request