基於階層式狄氏程序混合模型之音樂情緒標註之研究;A Study on Hierarchical Dirichlet Process Mixture Model Based Music Emotion Annotation

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/65803

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/65803

Title:	基於階層式狄氏程序混合模型之音樂情緒標註之研究;A Study on Hierarchical Dirichlet Process Mixture Model Based Music Emotion Annotation
Authors:	陳膺任;Ren,Chen-Ying
Contributors:	資訊工程學系
Keywords:	階層式狄氏程序;音樂情緒辨識;Hierarchical Dirichlet Process;Music Emotion recognition
Date:	2014-08-26
Issue Date:	2014-10-15 17:10:41 (UTC+8)
Publisher:	國立中央大學
Abstract:	音樂在人們生活中有著舉足輕重的地位，在資訊數位化且越來越容易保存的現代，音樂資料庫也更加龐大，因此需要一個自動化的分類方式，來幫助人們快速找到自己需要的音樂。傳統上可能會以歌手、類型來做為類別，但是音樂真正影響我們的其實是它所釋放的情感，因此以音樂情緒作為類別來進行標註或檢索的研究，也逐漸受到重視。傳統上的情緒模型，將每個類別分開建構，但真實世界中情緒的界定並不是那麼清楚，也就是情緒之間會有一些模糊地帶或是重疊。而我們考慮到這一點，並基於階層式狄氏程序混合模型(Hierarchical Dirichlet Process Mixture Model)可以共用成分的特性，建立模型間的連結關係，提出一個音樂情緒標註與檢索系統。同時我們也考慮到共用的特性可能會造成類別間的混淆，因此我們基於線性鑑別分析(Linear Discriminant Analysis) 的概念，在系統中加入了鑑別性的因素，另外，本系統以一個對應整體情緒成分的權重向量，來表示每個情緒類別。對於測試資料，我們也提出了三種不同的方法，來產生測試資料對應的權重向量。而我們利用該權重向量來判斷測試資料是否包含某個情緒。在實驗部分，其結果顯示我們提出的系統不論在標註或檢索上都有更好的表現，而我們也將討論不同的計算測試資料權重方法的差異。;The development of digital technology has enabled the storage of large collections music could be. For the convenience of users, some music database applications tag songs with some class labels. In tradition, music was classified by artist or genre, but the real influence of music is the emotion which it releases. Therefore, researchers have recently been studying the music annotation and retrieval method. In tradition, the model of each emotion was constructed individually, but an emotion cannot be defined clearly in the real world because the classes of emotions are usually considered overlapping. Accordingly, this paper proposes an music annotation and retrieval system that is based on hierarchical Dirichlet process mixture model (HDPMM), whose components can be shared between each model of emotions. Moreover, an improvement in HDPMM is proposed by added a discriminant factor to the proposed system based on the concept of linear discriminant analysis. The proposed system represents an emotion using a weighting coefficient that is related to a global set of components. Moreover, three methods are proposed to compute the weighting coefficients of testing data, and using the weighting coefficient to determine whether the testing data contain certain emotional content or not. Experimental results show that the proposed system performs well in automatic music emotion annotation and retrieval tasks. Finally, the evaluation of the three methods for computing weighting coefficients of testing data is also discussed in the experiments.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	511	View/Open

社群 sharing

Loading...