以基因演算法為基礎自動化產生特定情緒的音樂片段

Abstract

過去在自動化編曲的研究中，大部分的重點都專注在演算的過程上，如演算方法的創新、如何使樂曲更和諧等；近年來則開始從使用者的立場來設計演算法，像是如何產生出更符合使用者喜好的曲子。本研究希望結合演算式編曲及音樂情緒辨識(Music Emotion Recognition)來產生出具有特定情緒的音樂片段，也就是透過音樂情緒辨識中的音樂情緒模型用以分類情緒的特徵，做出帶有這些特徵的音樂並探討其中所帶給聆聽者的感受。本研究設定的音樂類型為帶有療癒感的背景音樂，也就是能夠使人感覺到放鬆心情及紓解壓力的音樂，透過網路上找到的相關音樂，分析當中的特徵並對應到音樂情緒模型上，最後透過問卷的方式來調查群眾對於演算產生的音樂的感受，探討是否達到本研究所想要做的帶有療癒感的音樂，並且請教音樂學者對於演算出來的音樂有何看法，透過學者的意見及群眾資料的比對，了解群眾及學者得意見是否一致及其中的差異。

Previous research of algorithmic composition mostly focus on the process of computing, such as developing new algorithms or how to make music more harmony. Recently, some research have tried to design their systems in a different point of view, which is how to make their music more favorable to user. This thesis aims to combine algorithmic composition with music emotion recognition, allowing our composition system to generate music with specific emotion. Our system composes music with some particular characteristics defined in music emotional models within many music emotion recognition systems. The music type selected in this thesis is background music which help people relax and release pressure in life. We call this kind of music relax music. We first analyze characteristics in relax music, and comparing them with characteristics defined in music emotional models. Secondly, we conduct an online survey to collect the opinions from ordinary listeners of the music generated by our system and whether they feel relaxed after listening to our music. Finally, we also ask music specialists about their opinions and suggestions to investigate the difference between specialist and ordinary listeners.