210 likes | 395 Views
Music Classification Using Significant Repeating Patterns. Chang-Rong Lin, Ning-Han Liu, Yi-Hung Wu, Arbee L.P. Chen, Proc. of 9th International Conference, Database Systems for Advances Applications (DASFAA 2004). Adviser:Jia-Ling Koh Speaker: Yu-ting Kung. Introduction.
E N D
Music Classification Using Significant Repeating Patterns Chang-Rong Lin, Ning-Han Liu, Yi-Hung Wu, Arbee L.P. Chen, Proc. of 9th International Conference, Database Systems for Advances Applications (DASFAA 2004) Adviser:Jia-Ling Koh Speaker: Yu-ting Kung
Introduction • Automatic classification of music data becomes a critical technique for providing an efficient and effective retrieval of music data • In this paper, propose a new approach for classifying music data based on their contents: • Rhythm • melody
Introduction (Cont.) • Moreover, use Repeating Patterns of music data to do music classification. • The repeating pattern that satisfy the constraints are called significant repeating patterns (SRP) • This approach contains two stages • Feature extraction • SRP-based classification
Classified music Music to be classified Feature extraction Generation of Significant repeating patterns SRP-based classification Class determination Relevance of source SRP Introduction (Cont.) • The flowchart of this approach Representation of music (melody & rhythm) Usefulness of SRP for classification Similarity measure for SRP matching Importance of Source SRP
Feature Extraction • Representations of Music • Rhythm: sequence of beats in music • According to the duration of a note, classify each note into one of the types in rhythm • Ex:
Feature Extraction (Cont.) • Melody: sequence of pitches in music • According to the length of a pitch interval, classify each pitch interval into one of the types in melody • Ex:
Feature Extraction (Cont.) • Generation of Significant Repeating Patterns • Adapt finding repeating patterns from a music piece by considering the following constrains: • Maximum Length • Minimum Length • Minimum frequency
SRP-based Classification • Usefulness of SRP for Classification • Fx,m: the frequency of the SRP x for the music m • Sup(x,m): the support of SRP x with respect to m • ASup(x,C): the aggregate support of SRP x
SRP-based Classification (Cont.) • NSup(x,C): the normalized support of SRP x • TS(x): sum up x’s normalized support in all classes • PW(x,C) : the pattern weight of SRP x in class C
SRP-based Classification (Cont.) PW(I,One) =1/(1+0.63) • Ex: Sup(V,D)=2/(2+3) ASup(VI,Two) =0+0.6 NSup(V,Two) =(1-0.4+1)/(1-0.4+1)
SRP-based Classification (Cont.) • Similarity Measures for SRP Matching • Given a Source SRP, adopt the dynamic programming approachto measure the similarity between it and each SRP in a class • Assign each symbol (beat or pitch) a numerical value
SRP-based Classification (Cont.) • Ex: • Beat • Pitch
(Rhythm: melody: SRP-based Classification (Cont.) • PS(x,y): the pattern similarity between two SRP’s x and y • E(x,C): Evidence Where y is the target SRP of x in C
For SRP X, the target SRP for class One is SRP III For SRP X, the target SRP for class Two is SRP I SRP-based Classification (Cont.) • Ex: • similarity threshold = 0.45, the music to be classified contains two source SRP’s X and XI Class One Class One E(X,ONE)=0.8x0.1=0.8, E(X,TWO)=0.6x0.39=0.234
SRP-based Classification (Cont.) • Class Determination • CS(C|m): classification score • Ex: Class One Class One
Experiment Results • Data Source • 7 classes: Blue, Country, Dance, Jazz, Latin, Pop, Rock music • Total music: 500 pieces of music • Training data: 4/5 of total pieces of music • Testing data: 1/5 of total pieces of music
Experiment Results (Cont.) • Impacts of Features • Rhythm: set minimum freq = 3 • Melody: set minimum freq = 2 • Sequence length is from 4 to 6
Experiment Results (Cont.) • Impacts of Similarity Threshold
Experiment Results (Cont.) • Comparison with the HMM-based Approach
Dynamic Programming Approach • The words `computer' and `commuter' are very similar, and a change of just one letter, p->m will change the first word into the second. • The word `sport' can be changed into `sort' by the deletion of the `p', or equivalently, `sort' can be changed into `sport' by the insertion of `p' • The edit distance of two strings, s1 and s2, is defined as the minimum number of point mutations required to change s1 into s2, where a point mutation is one of: • change a letter, • insert a letter or • delete a letter