DC Field | Value | Language |
---|---|---|
dc.contributor.author | Kim, Taejun | ko |
dc.contributor.author | Lee, Jongpil | ko |
dc.contributor.author | Nam, Juhan | ko |
dc.date.accessioned | 2019-06-12T07:50:19Z | - |
dc.date.available | 2019-06-12T07:50:19Z | - |
dc.date.created | 2019-06-12 | - |
dc.date.created | 2019-06-12 | - |
dc.date.created | 2019-06-12 | - |
dc.date.issued | 2019-05 | - |
dc.identifier.citation | IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, v.13, no.2, pp.285 - 297 | - |
dc.identifier.issn | 1932-4553 | - |
dc.identifier.uri | http://hdl.handle.net/10203/262580 | - |
dc.description.abstract | End-to-end learning with convolutional neural networks (CNNs) has become a standard approach in image classification. However, in audio classification, CNN-based models that use time-frequency representations as input are still popular. A recently proposed CNN architecture called SampleCNN takes raw waveforms directly and has very small sizes of filters. The architecture has proven to be effective in music classification tasks. In this paper, we scrutinize SampleCNN further by comparing it with spectrogram-based CNN and changing the suhsampling operation in three different audio domains: music, speech, and acoustic scene sound. Also, we extend SampleCNN to more advanced versions using components from residual networks and squeezeand-excitation networks. The results show that the squeeze-andexcitation block is particularly effective among them. Furthermore, we analyze the trained models to provide better understanding of the architectures. First, we visualize hierarchically learned features to see how the filters with small granularity adapt to audio signals from different domains. Second, we observe the squeeze-and-excitation block by plotting the distribution of excitation in several different ways. This analysis shows that the excitation tends to be increasingly class specific with increasing depth but the first layer that takes raw waveforms directly can be highly class specific, particularly in music data. We examine this further and show that the excitation in the first layer is sensitive to the loudness, which is an acoustic characteristic that distinguishes different genres of music. | - |
dc.language | English | - |
dc.publisher | IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC | - |
dc.title | Comparison and Analysis of SampleCNN Architectures for Audio Classification | - |
dc.type | Article | - |
dc.identifier.wosid | 000468435500009 | - |
dc.identifier.scopusid | 2-s2.0-85065982131 | - |
dc.type.rims | ART | - |
dc.citation.volume | 13 | - |
dc.citation.issue | 2 | - |
dc.citation.beginningpage | 285 | - |
dc.citation.endingpage | 297 | - |
dc.citation.publicationname | IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING | - |
dc.identifier.doi | 10.1109/JSTSP.2019.2909479 | - |
dc.contributor.localauthor | Nam, Juhan | - |
dc.description.isOpenAccess | N | - |
dc.type.journalArticle | Article | - |
dc.subject.keywordAuthor | Audio classification | - |
dc.subject.keywordAuthor | end-to-end learning | - |
dc.subject.keywordAuthor | convolutional neural networks | - |
dc.subject.keywordAuthor | residual networks | - |
dc.subject.keywordAuthor | squeeze-and-excitation networks | - |
dc.subject.keywordAuthor | interpretability | - |
dc.subject.keywordPlus | CONVOLUTIONAL NEURAL-NETWORKS | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.