實現於音訊壓縮域之內涵式歌者分類法

Design and Implementation for Content-based Singer Classification on Compressed Domain Audio Data

黃昱翔(Yu-siang Huang)

指導教授 ： 蔡宗漢

國立中央大學/資訊電機學院/電機工程學系/碩士(2010年)

摘要

在本論文中，我們提出了一個實現於MP3及AAC音樂壓縮域的自動化歌者分類法。不同於早年在MP3壓縮域使用MDCT (Modified Discrete Cosine Transform) 係數的作法，在本論文中我們是使用梅爾倒頻係數 (Mel-Frequency Cepstral Coefficients, MFCC) 當作辨識之特徵值。雖然梅爾倒頻係數經常用於音樂分類及語者辨識，但是這類的研究大多都不是在壓縮域中實現，因為梅爾倒頻係數無法直接由壓縮域中取得。在本論文中，我們使用了一個修正的梅爾倒頻係數計算法，使得梅爾倒頻係數可以從MP3及AAC音樂壓縮域中取得。除此之外，為了描述特徵空間中梅爾倒頻係數向量的分布，我們使用了高斯混合模型 (Gaussian Mixture Model, GMM)。而為了找出最相近的歌者/分類，我們則是使用最大似然分類法 (Maximum Likelihood Classification, MLC)。藉由最大似然分類法，每一個輸入的梅爾倒頻係數向量將會分配到其最相似的群聚中。最後，我們將演算法實現在兩個不同的嵌入式平台上，分別是Socle CDK及ITRI PAC Duo。最後的實驗結果也證實了我們所提方法的可行性。

關鍵字

壓縮域 ； 內涵式 ； 歌者辨識 ； 歌者分類

並列摘要

In this thesis we proposed a singer classification approach to automatically identify the singer of an unknown MP3 or AAC audio data. Differing from previous researches for singer identification in MP3 compressed domain, we use Mel-Frequency Cepstral Coefficients (MFCC) as the feature instead of MDCT (modified discrete cosine transform) coefficients. Although MFCC is often used in music classification and speaker recognition, it can not be directly obtained from compressed music data such as MP3 and AAC. In this thesis, we introduce a modified method for calculating MFCC vector in MP3 and AAC compressed domain. Besides, for describing the distribution of MFCC vectors in MFCC feature space, the GMM (Gaussian mixture model) is used. And then, for finding the nearest singer, we use maximum likelihood classification (MLC) to allot each input MFCC vector to its nearest group. Finally, we implement our approach on two embedded platforms, including Socle CDK and ITRI PAC Duo. Except the two embedded platforms, two operation system configurations are adopted, including Linux and Android. The experimental result verifies the feasibility of the proposed approach.

並列關鍵字

classification ； identification ； content-based ； compressed domain ； MP3 ； AAC

參考文獻

[1] Wang Y, Yaroslavsky L, Vilermo M (2000) On the relationship between MDCT, SDFT and DFT. Proceeding of the 5th International Conference on Signal Processing, vol 1, pp. 44-47

[2] Chang LY, Yu XQ, Wan WG, Li CL, Xu XQ (2009) Research and realization of speech segmentation in MP3 compressed domain. Journal of Computer Applications 29(4):1188-1192

[3] Logan B (2000) Mel frequency cepstral coefficients for music modeling. Proceeding of the 1st International Symposium on Music Information Retrieval.

[4] Langlois T, Marques G (2009) A music classification method based on timbral features. Proceeding of the 10th International Society for Music Information Retrieval conference, pp. 81-86

[6] Liu CC, Huang CS (2002) A singer identification technique for content-based classification of MP3 music objects. Proceeding of the 11th International Conference on Information and Knowledge Management, pp. 438-445