Font Size: a A A

Research On Chinese Digit Speech Recognition System Based On HTK

Posted on:2009-10-28Degree:MasterType:Thesis
Country:ChinaCandidate:M H ZhongFull Text:PDF
GTID:2178360245459612Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
With the lasting development of computer and information technology, speech interface would become an indispensable human-computer interactive component. The speech recognition(SR) technique has been developed about 50 years, getting extensive applications such as speech-dialable telephone, speech-controllable electronic appliances and speech- controllable industry devices, but far to perfect to the extent that none of any further research and further improvement deserves engaging in. The technique has also come out of laboratory and infused new high-tech element into more and more electronic products. However, as simple as the Chinese digit SR at first glance, there are still numerous subjects deserving further intensive researches. This thesis describes researches of the prototyping of a HMM-based digit SR system in MATLAB environment, and then the implementation of C-code sources based on HTK(HMM toolkit), for both of Chinese isolated digit and continuous digit SR systems.Firstly, this thesis introduces the state of the art SR recognition techniques and some difficultices in the implementation of Chinese continuous digit SR system, further gives related background materials and the purpose of this research.Secondly, the principles for the modeling and construction of SR system are detailedly explained, including speech mathematical models, speech signal-processing method, feature extraction method, and the choice of MFCCs as speech features. After that, HMM-based SR algorithm, used in this thesis, are expounded, including the definition of HMM, three fundamental SR problems and primary algorithms. Then, argumentation stress is changed to the experimentation side problems of HMM, mainly the optimal numbers of states and mixture in HMM.Thirdly, HTK Software Architecture and HMM Toolkit are introduced briefly. And implementation process of the SR system based on HTK is detailedly outlined. By experiments of isolated digit and continuous digit SR systems, it is clear that correct configuration of suitable recogised unit, right Gaussian mixture component number and proper MFCCs dimensions is justified for the improvement of Chinese SR system.Finally, some conclusions are derived for the implementation of Chinese digit SR system and advices for further research are also given.
Keywords/Search Tags:Speech Recognition(SR), Feature Extraction, Hidden Markov Model(HMM), HTK, Recognition Unit
PDF Full Text Request
Related items