基於樂譜中音樂術語及其他影響因素之分析驗證音樂情緒軌跡追蹤系統

A verification to the trajectory of music emotion tracking system by musical terms in the score

徐斐(Hsu, Fei)

指導教授 ： 鄭泗東

國立交通大學/工學院/聲音與音樂創意科技碩士學位學程/碩士(2020年)

摘要

本研究使用音樂情緒軌跡追蹤程式，從不同的角度、類型，配合樂理，進而分析古典音樂在情緒上的變化，並在各類的古典曲風上做出比較。音樂情緒軌跡追蹤程式結合類別式情緒分類法及二維情緒平面座標作為情緒辨識的基礎模型，配合機器學習技術與音樂訊號的處理，建立即時性音樂情緒軌跡追蹤系統，將音訊所引發的情緒以視覺化的方式呈現在結果之中。音樂情緒軌跡追蹤系統採納四種不同的情緒作為辨識的基礎，並使用高斯混合模型(GMM)作為分類四種情緒邊界的手段。
基於”Pleasant”、”Solemn”、”Agitated”、”Exuberant”四種基礎的情緒分類，從中分析音色、音量、音樂事件密度、和聲不和諧度、調式等音樂的基礎特徵。古典音樂的作曲家在創作時，常藉由音樂術語及聲音變化量的符號來描繪曲子，而這些因素往往是影響整首歌曲情緒起伏變化的主因。至今已有諸多研究在心理、生理上，對歌曲進行情緒上的分析；本研究則使用程式將音樂情緒數據化，以樂理為輔，得到較為準確的古典音樂情緒分析， 與心理、生理之學說互相呼應。

關鍵字

音樂情緒辨識 ； 高斯混合模型 ； 古典音樂分析

並列摘要

This study uses the music emotion trajectory tracking program to analyze the emotional changes of classical music from different ways and types, and to compare them in various classical styles. The music emotion trajectory tracking program combines the category emotion classification method and the two-dimensional emotion plane coordinates as the basic model of emotional recognition. This system cooperates with the processing of machine learning technology and music signals to establish an instant music emotion trajectory tracking system, which will bring the emotions triggered by the audio to The visual approach is presented in the results. The music emotion trajectory tracking program adopts four different emotions as the basis for identification, and uses the Gaussian mixture model (GMM) as a means of classifying the four emotional boundaries. Based on the four basic emotion categories, namely, "Pleasant", "Solemn", "Agitated", and "Exuberant", the basic characteristics of music, such as tone, volume, music event density, harmony dissonance, and tonality are analyzed. The composers of classical music often use the symbols of musical terms and sound changes to describe the songs, and these factors are often the main causes of the emotional ups and downs of the whole song. So far, there have been many studies on the psychological and physiological aspects of the emotional analysis of songs; this study uses the program to analyze music emotions, supplemented by music theory, to obtain more accurate analysis of classical music emotions, which responds to the study of psychology and physiology.

並列關鍵字

Music Emotion Recognition ； Gaussian Mixture Model (GMM) ； Classical Music Analysis

參考文獻

1.傅俊傑，「具時變情緒軌機介面之自動音樂情緒追蹤系統」，國立交通大學聲音與音樂創意科技碩士學位學程，碩士論文，民國九十九年

2.Feng, Y., Zhuang, Y., Pan, Y., “Music information retrieval by detecting mood via computational media aesthetics,” Proceedings of the 2003 IEEE/WIC International Conference on Web Intelligence Washington, DC, USA 2003.

3.W. Chai and B. Vercoe, “Detection of key change in classical piano music”, ISMIR, London, 2005.

4.L.Lu, D.Liu, and H.-J.Zhang, “Automatic mood detection and tracking of music audio signals,” IEEE Transactions on Audio, Speech, and Language Processing, vol 14 No 1,: pp. 5-18, January 2006.

5.K Trohidis, G Tsoumakas, G Kalliris, “Multilabel classification of music into emotion,” ISMIR, 2008.