선택한 단어 수는 입니다.
최소 단어 이상 선택하여야 합니다.
최소 단어 이상 선택하여야 합니다.
선택한 단어 수는 30입니다.
최대 10 단어까지만 선택 가능합니다.
최대 10 단어까지만 선택 가능합니다.
제안 방법
- The OSC was first proposed solely for musical genre classification and describes the difference between the maximum and the minimum of the power spectrum at the octave-scale subbands. As an extension to OSC, this paper investigates different types of the spectral distributional characteristics of the octave-scale subbands. In particular, we propose a musical genre classification method based on the octaveband order statistics, such as the median, quartile, minimum, maximum, and so on.
- The short-time spectral features in the proposed method are based on the distributional characteristics of the octavescale subbands. According to the results in the paper,[6] the octave-scale subbands contain enough information for distinguishing the genres of a music signal.
- The mean and the standard deviation of the frame-level features in a segment are widely-used as a segment-level feature for most of the previous works. In this paper, we also extend the previous temporal integration methods, the mean and the standard deviation, into the order statistics. We apply the same types of the summary order statistics in Table 1 to temporal integration of framelevel features.
- The extracted short-time features were temporally integrated over six seconds. Then the linear SVM classifier was trained and tested in classifying a segment-level feature. The genre of each music clip was determined by the majority voting on the classification results of the segments in the clip.
- The classification results of the ISMIR2004 and GTZAN datasets are given in Table 2 and 3 respectively. The set of octave-band order statistics, SOS1, SOS2, SOS3, SOS4, and SOS5, in Table 1 was used in combination with the temporal integration methods, TOS1, TOS2, TOS3, TOS4, and TOS5. For a comparison with the previous spectral descriptors, the 12-order MFCC and the OSC were included in the test.
- We note that it is reported in[2,9] if they use the SVM with RBF kernel, the classification accuracy can be improved further about 5 %. Since an extensive benchmark testing is not the aim of this paper, we focus on showing the validity of using the octave-band order statistics on the musical genre classification task. By using the SOS5 and TOS5, the classification accuracy of the proposed method exceeded 84 % on both datasets in Tables 2 and 3, which is among the best results reported so far with the linear SVM classifier, although the proposed method is less complicated than the other approaches compared with.
대상 데이터
- The second music dataset (abbreviated as GTZAN) is the one that was used by George Tzanetakis in his work.[1] It consists of 1000 songs over ten different genres: blues, classical, country, disco, hiphop, jazz, metal, pop, reggae, and rock. For the ISMIR2004 dataset, one half of the songs was used for training, and the other half was used for testing.
- [1] It consists of 1000 songs over ten different genres: blues, classical, country, disco, hiphop, jazz, metal, pop, reggae, and rock. For the ISMIR2004 dataset, one half of the songs was used for training, and the other half was used for testing. For the GTZAN dataset, the 10 fold cross-validation was used to get the classification accuracy.
이론/모형
- Any kind of state-of-the-art statistical classifiers, such as nearest neighbor, Gaussian mixture model, and support vector machine (SVM), can be used in training the genre model over the segment-level feature. In this paper, the linear SVM classifier, known for its simplicity and reasonably high classification accuracy, is used. As a final step, the classification results from all segments in a music clip are aggregated typically by the majority voting rule.
- For a comparison with the previous spectral descriptors, the 12-order MFCC and the OSC were included in the test. To compare temporal integration methods, the temporal mean and standard deviation (abbreviated as TMS) was included in the test. The results of them are also listed in Tables 2 and 3.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.