| Summary: | 碩士 === 國立東華大學 === 資訊工程學系 === 102 === The paper presents the design of a Music Melody Plagiarism Detection (MMPD) system to locate and detect the possible plagiarism in songs. Our approach is to devise a novel algorithm to find the similar segments of music melodies between two songs. The algorithm is based on the Warping Time Series Join (WTSJ) algorithm which was originally designed to search the appropriate points for smoothly joining two motion sequences. Our work extends the WTSJ algorithm to address the problems not considered in the original design. Besides, the target application switches from the analyses on motion sequences to that on music melodies, suffering from some other issues in practical considerations.
The major contributions made in this thesis include the following:
1. designing a Noise-Blocked Path Connecting method to tolerate noisy disturbance caused by inaccurate tone extraction in music melodies,
2. designing a Distorted Subsequence Correspondence Checking method to find the detailed alignment between the music tones of spotted similar melody segments,
3. designing an Overlapped Path Pruning method to remove redundant detection of similar melody segments,
4. designing a Distorted Path Pruning method to avoid the false detection of melody segments with inappropriate tone alignments, and
5. designing a Length-Limited Pruning method to remove the false detection of short melody segments.
The MMPD system is also capable of performing the function of “Query by Humming” which is very useful for the music retrieval in music databases. The performance of the system is evaluated through a number of experiments. The results show that the designed system achieves better performance than other existing methods.
|
|---|
