MP3音樂的聆賞情緒自動分類

王鴻文 ； 劉志俊

《Journal of Information Technology and Applications(資訊科技與應用期刊)》 4卷4期 (2010/12) Pp. 160-171

https://doi.org/10.6302/JITA.201012_4(4).0001

摘要

本篇論文探討音樂給人在感知方面的高階情緒特性，並據此提出一個音樂情緒模型。我們希望能針對MP3格式的數位音樂，透過擷取出音樂原始的低階特徵，來計算與分析音樂感知相關高階特徵。速度、調性、力度被認為是影響音樂表情的三種主要因素，因此我們針對MP3格式的音樂提出兩種方法來自動偵測上述因素之值。接著根據我們所提出的情緒模型，將這三種音樂因素轉換成三維聆賞情緒模型，並對應至Hevner所建議的八種情緒。透過我們所提出的高階音樂感知特徵計算與音樂情緒模型，我們可以自動判別一首未知MP3音樂的聆賞情緒之類型。由於音樂本身本質是聽覺的媒體，在很多狀況下，人們所感受的音樂聆賞情緒，並不是單一的、彼此無交集的（disjointed）。所以一首歌曲的情緒分類結果應為八種基本情緒的個別傾向程度的組合。此外，聽覺的媒體在許多應用上希望能以視覺化的（visual）方式來呈現。因此，我們試著將一首音樂給人的感受，利用顏色這種視覺化的方式，來呈現聽覺性、看不到、較抽象的音樂聆賞情緒。為了以圖型視覺方式來呈現音樂給人的各種情緒比重，我們提出一個以音樂情緒色彩對應雷達圖來表現音樂情緒與色彩的關係。

關鍵字

多媒體內涵式分析 ； 音樂聆賞情緒 ； 人類感知 ； 音樂情緒模型 ； 音樂內容視覺化

並列摘要

Far before any forms of verbal language emerged, human beings have learned to express their thoughts and feelings through vocal variations in tone and force. With the coming of the digital era, the application of digital multimedia data have been increasing, and content-based multimedia analysis has become the focus of recent research. Former content-based multimedia analysis focused mainly on low-level signal analysis. Recent analysis, with great progress, has turned to center on high-level human perceptional and psychological analysis.The purpose of this paper is to propose a musical mood model by studying the high-level emotional features that music has been bringing to human beings. Aimed at MP3 digital music and featuring primary low-level musical characteristics, we try to analyze perception-related high-level characteristics. As tempo, dynamics and key are believed to be the three main factors in influencing musical expression, we propose two approaches which will automatically detect the above factors in MP3 music. Then, these three factors will be transformed into three dimensions in the proposed emotional model, and combined and arranged so that they correspond to the eight mood classifications suggested by Hevner. By referring to these high-level perceptional features and the musical emotional models we have proposed, we will be able to automatically classify moods in MP3 music.As music is in essence the media of hearing, perceptions, in many circumstances, are not sole or disjointed. The result of mood classification in a song should be the combination of eight emotional tendencies. Furthermore, audio media are often expected to be represented visually. In this paper we try to present mood that music has brought to the hearer, the acoustic, visible and abstract emotions, by way of color. Hence, in order to the proportion of each mood classification that music brings, we propose here a radar diagram showing the correspondence between musical mood and color.

並列關鍵字

content-based multimedia analysis ； human perception ； music emotion model ； music visualization