Reconstruction Algorithms For Compressive Sensing And Their Applications To Digital Voice Recognition

Language is the most important means of communication. It is convenient andcoherent, efficient and accurate. With the constant development of society and science,a variety of machines participate in the human activities and social events, so dealingwith the relationship between machines and people, and making people manipulatemachines more easier is getting more important. With the extensive application ofcomputers and artificial intelligence machines, people have learned the languagecommunication between man and machine is the most appropriate means ofcommunication, and its acoustic manifestation of language are speech sounds. If youwant the machine to understand human words and to speak out, these will require a lotof work to be done. This work is studied by scientists for decades before they foundspeech recognition and speech synthesis technology. With the rapid development ofmobile communication technology, people can communicate to anyone by telephoneat any time and any place, and voice compression coding technology plays a key roleduring this procedure. These applications become the main content of the speechsignal processing.Speech signal processing is an interdisciplinary subject which combines digitalsignal processing technology with phonetics, it has a close link with psychology,computer science, cognitive science, pattern recognition, artificial intelligence,linguistics and other disciplines. The development of speech signals processingtechniques rely on these subjects, and the advancement of voice signal processingtechnology also promote the progress of these disciplines.In this paper, when using the basic principle of compressed sensing toreconstruct the speech signal, in view of their own characteristics of the orthogonalmatching pursuit (OMP) algorithm and the subspace tracking (SP) algorithm, andcombining with self-related thoughts, an improved sub-autocorrelation blocksubspace tracking (BASP) algorithm is proposed. The experimental resultsdemonstrate that the recovered speech signal resulting from BASP method has bettersignal to noise ratio, and high MOS score than the Subspace Pursuit algorithm in thesame compression ratio. Then in this paper, in view of features that the method of lpcc has weakanti-noise property when Conducting voice recognition, combining with features ofcompressed sensing that it can recover large amounts of information when basing on asmall amount of information, this paper proposes a method of voice recognitionsystems which combine lpcc and compressed sensing, in order to combinecompressed sensing methods and lpcc method, and this paper also modifiedreconstruction process of compressed sensing, proposing an adaptive threshold toconstrain the reconstruction process. Experimental results show that the improvedalgorithm can effectively improve the anti-noise property of the system.