Font Size: a A A

Research On Audio Signal Classification Based On Tensor Decomposition

Posted on:2021-05-09Degree:MasterType:Thesis
Country:ChinaCandidate:W C XinFull Text:PDF
GTID:2428330629982586Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the continuous progress and development of multimedia and Internet technologies,audio signals,as an important component of multimedia signals,are favored by more and more researchers for information processing and mining,and there are various methods for processing these audio signals.With the popularity of the Internet,when we select useful information from a large number of audio information,we can get rid of the information that is useless and affects researchers to complete audio classification,which not only saves human resources,but also enables us to extract useful information more efficiently and conveniently and screen the redundant information.Therefore,in this age of information mixing,audio classification technology has great potential.Under the background of artificial intelligence and big data analysis,image audio and other fields have shown a relatively bright future,and audio scene classification has gradually become the favorite direction of researchers.Audio scene classification is analyzed by audio label information.Compared with the previous single camera for scene classification,the audio technology does not need to take into account such bad weather factors as fog and heavy rain,and the dead Angle problem like video surveillance will not exist in the process of audio technology acquisition.Audio classification only requires a device to collect sound and a device to receive it.Video monitoring,if combined with audio technology,will have a positive impact on our production and life in the future,so that we will have a broader space for development in the field of scene classification.Faced with a series of classification problems,this project USES tensor analysis to construct audio tensors,and then carries out tensor decomposition,so that sound features with strong robustness can be obtained,and the classification accuracy can be improved.Tensor analysis is a technique which USES the form of tensor to represent the signal and the correlation operation of tensor to analyze and process the signal.Compared with the traditional methods of vector and matrix analysis,tensor analysis can make full use of the correlation among various factors,so as to analyze and operate the audio signal as a whole in the higher order space.In this study,the third-order tensors are used as the audio features,and the support vector machine is used as the classifier to complete the audio scene classification.The scenes included 10 categories: car horn,engine idling,gunshot,children's game,dog barking,and street music.The data volume of the whole experiment was 8,732 audio clips.Finally,the accuracy of classification was 91.3%,and the accuracy of single scene classification reached more than 80%.This shows that the audio classification method studied in this paper is appropriate,which lays a foundation for further research.
Keywords/Search Tags:Audio classification, Feature extraction, Tensor analysis, Mel frequency cepstrum coefficient, Support vector machine
PDF Full Text Request
Related items