以音度與節奏為導向之音樂拼貼

Abstract

在我們的論文之中，我們提供了一個從給定音樂中，自動選取合適音樂片段的系統，並且盡可能地把選取出來的音樂片段順暢的串接起來。然而如我們所能理解，要拼接不同音樂片段且不導致任何瑕疵是一個十分棘手的問題。主要的難度在於人耳的聽覺跟電腦所感應的聽覺，實際上是有很大的差距的。此外，結合起來的結果好壞與否，其實主要是取決於人們的感受。於是我們對拼貼方法的關係以及拼接音樂的聽覺品質理想與否作了一連串的實驗來佐證我們的假設。整體而言，我們所拼貼出來的音樂，大多能被受試者所接受，有時甚至會讓他們驚呼不已。所以有了這樣的系統之後，對於多媒體聲音檔案有諸多的應用，例如像是自動產生擴張或縮減的背景音樂，或者是自動產生舞蹈組曲。

In this thesis, we provide a tool for automatically choosing proper music clips from a given audio clip collections and combining the chosen clips together as euphonious as possible. To seamlessly concatenate two different music clips without causing any audible defect is really a hard nut to crack. The main difficulty lies in the gap between human feeling and computer hearing. Since whether the combined music euphonious or not highly depends on human’s feeling, we conduct subjective evaluation on the correlation between pasting methods and auditory quality of combined clips. The experimental results show that the automatically generated music pastes are acceptable to most of the evaluators. The proposed system can be used to generate lengthened or shortened background music and dancing suite, which is useful for some audio-assisted multimedia applications.