人工智慧生成之歌曲的音樂風格研究

Musical Style in Songs Generated by Artificial Intelligence

鑒於運用人工智慧生成流行音樂（popular music）作品的技術正方興未艾，卻少有研究以嚴謹的分析評估其生成內容之品質，因此，本研究從音樂風格特徵切入，探討程式所生成歌曲與人類音樂家作品的關係。本研究基於爵士（jazz）、搖滾（rock）、嘻哈／饒舌（hip-hop／rap），分析程式Suno（v4）依據這三個詞彙所生成之音檔，以及文獻中描述上述類型音樂之特徵，並且比對樂曲分析與文獻分析結果。本研究發現程式生成之音樂風格與2010年代以後的流行（pop）歌曲大致相近，但尚具以下差異：其一，曲式較為複雜、不對稱，且有並置不同年代典型聲響之傾向；其二，程式能成熟運用多項作曲和音樂工程方面之技術，但難以掌握這些手法與音樂以外之內容的連結與配合。此外，基於分析過程中所梳理出人工智慧對大量過往人類樂種的風格化行為，及其目前所呈現的聲響特徵，本文認為可視之為流行（pop）作為獨立樂種於當代表現出的一種獨特風格。並且，建議音樂家和音樂學者應密切觀察與研究，這種風格是否會在未來，對主流聽眾的審美產生顯著影響。

Although generative artificial intelligence in popular music has been developing rapidly, few studies have applied rigorous analytical methods to evaluate the quality of AI-generated content. Therefore, this study focuses on musical traits and explores the stylistic differences between songs generated by AI and those created by human musicians. The author compares audio files generated by Suno (v4) using the genre labels“jazz,”“rock,”and“hip-hop/rap”with the characteristics of these genres as described in the literature. This study finds that works generated by AI are stylistically similar to pop songs produced after the 2010s, but several differences remain: firstly, the generated forms tend to be more complex and asymmetrical, and the AI system shows a tendency to juxtapose characteristic sounds from different historical periods; secondly, in terms of music composition and engineering, the AI employs a variety of techniques but can barely grasp the coordination between these techniques and extra-musical content. Furthermore, considering the stylizing behavior of state-of-the-art AI across a large number of musical genres from the past and the sound characteristics that result from this process, its unique style may be regarded as a distinctive style in contemporary pop genre. Uncertainties remain about whether this style will have a significant impact on the aesthetics of mainstream audiences, which makes it a valuable topic for future research and observation.