Music Matching Speaker : 黃茂政 指導教授 : 陳嘉琳 博士

? Introduction : Input Short piece of music Music Database Output Candidate songs

Outline : Introduction Music Databases : Indexing Techniques and Implementation Pattern Recognition Conclusion

* music object : MIDI , WAV , other types of media Introduction : * music object : MIDI , WAV , other types of media associated with music. * requirements : unstructured search. Input fault tolerance. (duplication, elimination, disorder, irrelevant) Saving of storage space and high performance.

MIDI vs. WAV format : WAV file information : Channels, Sampling, Bits, Sound data

Music Databases : Indexing Techniques and Implementation Arbee L. P. Chen Chord-representation model. Theme consists of a mdelody and a thythm. * melody : sequence of the pitches of all notes. * rhythm : sequence of the durations of all notes. * chord : a signature of a melody.

Music Databases : Indexing Techniques and Implementation Frequently-used chords C : do, mi, sol Dm : re, fa, la Em : mi, sol, si G : sol, si, re Am : la, do, mi Example : measure with six notes : do, mi, mi, sol, sol, sol “ C “ chord.

Music Databases : Indexing Techniques and Implementation Using PAT-tree as Index Structure Text : ababc substring : ababc, babc, abc, bc, c ab b c abc c abc c ababc abc babc bc c Operation : Insert, Delete, Search.

Music Databases : Indexing Techniques and Implementation Problem : 1. The chord-set decides the search performance. 2. When the users mistakenly rise the pitch to higher or lower key, the search operation will repeal. “We must alter the representation model to solve this problem.”

Pattern Recognition : MIDI curve feature A/D shift Candidate songs (sampling) Critical value shift Normalized crosscorrelation Candidate songs

Sampling : 精確 與 效率 trade off

Index l is the shift parameter. Normalized crosscorrelation : x : The (input) pattern. y : The reference pattern or template for a particular class. r : The extent of x over which the match occurs. Index l is the shift parameter.

Normalized crosscorrelation : <example>

chord-representation Conclusion : chord-representation pattern recognition unstructured search v x fault tolerance * v storage space v v performance ? ? http://140.126.11.7/ http://db.nthu.edu.tw / Acknowledge : 蘇文鈺 老師