| Endpoint detection is to accurately identify the start point and the termination point ofthe speech signal, from a specified period of the speech signal, the endpoint detection is avalid voice signal and the unwanted noise signal to distinguish. Speech endpoint detection inthe past, have been proposed a variety of methods. But, these methods are there such asdefects and deficiencies, voice processing, the final results are not very satisfactory. Thisarticle summarizes some excellent ideas of their predecessors on attempts to propose a newalgorithm-voice endpoint detection algorithm based on information entropy.Firstly, This paper begins with a brief introduction and analysis of the methods and ideasproposed in recent years, domestic and foreign research scholars voice endpoint detectionalgorithm based, a simple comparison of the advantages and disadvantages of thesealgorithms is proposed, the existing endpoint detection algorithm on the signal-to-noise thanhigh voice endpoint detection effect is good, and the algorithm is simple and practical, butwhen there is background noise, the performance has a significant degree of decline, someeven lapsed. Voice endpoint detection algorithm currently used comparative study of the basicconcepts of the speech signal, such as the information signal, the information entropy and itsrelated content simple summary and introduction. Deal to directly with the input voice signalthat there will be a lot of inconvenience and error caused by the direct effect of the treatmentis not ideal, so often voice signal preprocessing work before making real voice informationprocessing.The article next on the pre-processing of the speech signal detail. Including voice signalacquisition, digitization of the signal, and sub-frame plus window processing, as well as howto select a window function and several voice signal processing met the common windowfunction in all aspects of the time domain and frequency domain, the low-pass and high-pass,etc. The properties were compared.Next, the article introduces the concept of entropy related. Including the origin of theinformation entropy, the definition of the algorithm as well as the physical significance of theinformation entropy, information entropy basic properties, and applications of informationentropy simple summary and outlook. After the article on voice endpoint detection algorithmbased on information entropy detailed theoretical analysis and formula derivation. The entire speech endpoint detection processes and steps to verify the feasibility of voice endpointdetection algorithm based on information entropy in theory.At last, It has verified the mathematical derivation of the formula to writeapplications, and design entropy spectral function in the MATLAB simulation platform.Recorded after various stages of information entropy, entropy function and discuss thejudgment before the threshold set, and then based on the information entropy predefinedthreshold comparison in order to determine the voice signal spurt state or the silent state,determine the start and end points of the speech signal. And intuitive drawing functionMATLAB simulation software mark the start and end points and end points of the speechsignal.Finally, through the establishment of reasonable voice processing and entropy functionand MATLAB simulation software actual simulation experiments detailed simulation imageanalysis and repeated values. Last found that when the matter of the external environment isno background noise, voice endpoint detection algorithm based on information entropy can begood to distinguish between speech segments and non-voice segment, by the voice of theenergy impact is relatively small and have a certain Robust.The experiment, however, still exist some shortcomings and it was found that in a highersignal-to-noise ratio based on entropy algorithm showed good robustness However, when thesignal-to-noise ratio is reduced, the robustness of the algorithm’s performance is notsatisfactory, the article also discussed and explained at last. |
PDF Full Text Request