| Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 96 === It is a non-trivial task for untrained people to appreciate classical music pieces. Thus, we present a method of music structure analysis in order to help people understand the conveyed messages in the music. Previous works in music structure analysis can be classified by the two types of input sources: symbolic representations (scores or midi files) and waveform format (mp3, wav, etc). The methods used in analyzing symbolic representations cannot be easily applied to waveform-format analysis due to the immaturity of polyphonic transcription technique. Accordingly, most researchers analyze the waveform-format music by recognizing the repetitive patterns represented in low level features extracted directly from the waveform. However, in this methodology, the importance of music theory is often neglected. Therefore, in this thesis, we focus on analyzing waveform-format classical music and attempt to segment music based on more in-depth music theory, such as the cadence. Moreover, instead of relying on the low level repeating structure, we address the structure analysis problem from the semantic perspective of “musical form”.
|
|---|
