iPlayr - 情緒感知音樂平台

Abstract

隨著音樂技術的發達,如今製作一首歌曲的門檻和難度已經非常低。也因此充斥在這個世界上的音樂變得玲郎滿目,要尋找一些自己想要聽的歌,常常無從找起。 一首歌曲包含有許多的資訊和感覺。所謂資訊包括聆聽者欣賞時的旋律、音訊;可閱讀的歌詞;一些背景資訊像是演奏者、歌名、專輯名稱、年份、音樂類型等,這些都可以說是一首歌所擁有的資訊。當然人與人之間所分享的知識也包括在其中。有人提供歌詞含意;有人分享心得;有人對歌曲下標籤;有人提供評分。這些也都是在做音樂分析上很重要的資源。 本論文延續了音樂分析的精神,專注於研究歌詞與歌曲背景對音樂情緒分析的影響,進而證明此資訊的重要性。這些資訊大部份以文字為主,因此本論文使用常識運算(Commonsense Computing)解決歌詞與歌曲背景的關鍵字擷取與情緒轉換的問題。並在論文最後實作iPlayr,一個提供基於情感及關鍵字搜尋的音樂平台。 再根據實驗結果,說明歌詞與歌曲背景的重要性。並在最後提出未來關於使用者與音樂可以有的互動與延伸。

With the development of the music technique, producing a song becomes much more easier. Consequently, it makes the songs existing in this world become more and more or even uncountable. If the user wants to search some particular songs that fit the user's feeling, it would be a tough task. Each song contains lots of information and feelings. Generally, the content of the song such as the melody or the lyrics, or the context like artist name, title, album, genre, etc. are defined to be the essential elements and information of a song. In this thesis, we focus on the textual and contextual analysis of music to verify their importance in music emotion analysis. We utilize Commonsense Computing to solve the emotion and keyword extraction problems. Based on these analyses, we build up an emotion-aware music platform, iPlayr, to demonstrate our ideas. Moreover, according to the evaluation results, we explain how these textual and contextual information affect the completeness and accuracy of music emotion analysis.