| Paper: | MLSP-P4.10 | ||
| Session: | Machine Learning Applications | ||
| Time: | Thursday, May 20, 09:30 - 11:30 | ||
| Presentation: | Poster | ||
| Topic: | Machine Learning for Signal Processing: Signal detection, Pattern Recognition and Classification | ||
| Title: | CONTENT-BASED MUSIC SIMILARITY SEARCH AND EMOTION DETECTION | ||
| Authors: | Tao Li; University of Rochester | ||
| Mitsunori Ogihara; University of Rochester | |||
| Abstract: | This paper investigates the use of acoustic based features for musicinformation retrieval. Two specific problems are studied: similaritysearch (searching for music sound files similar to a given musicsound file) and emotion detection (detection of emotion in musicsounds). The Daubechies Wavelet Coefficient Histograms (proposed byLi, Ogihara, and Li), which consist of moments of the coefficientscalculated by applying the Db8 wavelet filter, are combined with thetimbral features extracted using the MARSYAS system of Tzanetakisand Cook, to generate compact music features. For similaritysearch, the distance between two sound files is defined to be theEuclidean distance of their normalized representations. Based on thedistance measure the closest sound files to an input sound file is obtained. Experiments on Jazz vocal and Classical sound filesachieve a very high level of accuracy. Emotion detection is cast as a multiclass classification problem, decomposed as a multiplebinary classification problem, and is resolved with the use ofSupport Vector Machines trained on the extracted features. Ourexperiments on emotion detection achieved reasonably accurateperformance and provided some insights on future work. | ||
| Back | |||
Home -||-
Organizing Committee -||-
Technical Committee -||-
Technical Program -||-
Plenaries
Paper Submission -||-
Special Sessions -||-
ITT -||-
Paper Review -||-
Exhibits -||-
Tutorials
Information -||-
Registration -||-
Travel Insurance -||-
Housing -||-
Workshops