Font Size: a A A

Voice Features Short Statistical Analysis And Application

Posted on:2015-10-29Degree:MasterType:Thesis
Country:ChinaCandidate:L ChenFull Text:PDF
GTID:2298330434465303Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Voice as acoustic expression form of language, is a human basic carrier ofinformation communication between each other and important means. In speechsignal processing, the most important is can accurately extract the speech signalparameters. Because only obtain accurate, the essential feature of the speech signalcan be characterized parameters, is likely to use these parameters for efficient andreliable speech recognition, speech coding and speech compression codec processing,etc. Among them, the clear voiced judgment and pitch period of speech signal isparticularly important, the extraction of the sentence and extraction of accurate or notwill directly affect the speech synthesis are true, will be able to reproduce the originalspeech signal spectrum. With the rapid development of modern information scienceand technology, especially the increasingly perfect and popularization of computernetwork technology, makes the speech signal processing technology is playing anincreasingly important role in the contemporary era. At present, the speech signalprocessing technology and its application technology has become an important andindispensable component in the information society, have profound meaning topromote the development of the society in the future.Clear and dullness voiced judgment is one of the most basic question in speechsignal processing. At present, about the decision algorithm of clear voiced speechsignals have a lot of kinds, based on the quantitative clear dullness of recursiveanalysis, based on phase space reconstruction of voiced judgment, based on nonlinearvoice clear and clear voiced judgment of support vector machine, and voicedjudgment algorithm optimization in the vocoder, within the scope of these algorithmsin different application fields have significant effects. But the characteristics of thesealgorithms is that the computational complexity is relatively low, the accuracy of thejudgment is relatively low. And of all the algorithms, clear the voiced judgment byusing threshold, the choice of parameters can also affect the effect of decision. Aimedat the disadvantage of these algorithms, this paper proposes a new ruling voiceddetection method.This method is mainly for short-term features of speech and statistical analysis to estimate the probability by the prior probability of the estimated speech signal and theclass conditional probability density estimates derived a posteriori probability of thespeech signal estimate to be cleared using the Bayes formula voiced verdict. With theabove conclusion and then the voice signal pitch detection and formant estimates. Onthis basis, it is studied based on linear prediction (LPC) pitch detection methods, testresults can be seen through the system, these two methods can achieve detection ofthe pitch, with some effect and feasibility.
Keywords/Search Tags:speech short features, probability estimates, Bayesformula, pitch detection, LPC
PDF Full Text Request
Related items