Loading...
Please wait, while we are loading the content...
Similar Documents
Music Classification with the Munich Opensmile Toolkit ( Mirex 2010 Submission )
| Content Provider | Semantic Scholar |
|---|---|
| Author | Eyben, Florian Schuller, Björn W. |
| Copyright Year | 2010 |
| Abstract | We present the Munich openSMILE Music Classification system for the MIREX 2010 evaluations. The system is designed for three MIREX train/test tasks: US Pop Genre Classification, Latin Music Genre Classification, and Audio Mood Classification. The system is based on a feature set which combines 87 rhythmic features (including so called Tatum and meter vectors) with general spectral, energy, and timbre features. Statistical functionals (feature summaries) are applied to the spectral, energy, and timbre features, and the resulting 1,236 features are combined with the 87 rhythmic descriptors. The resulting 1,323 dimensional vector is classified with Support Vector Machines which are trained via Sequential Minimal Optimisation. MIREX evaluations are conducted in a 3-fold Cross Validation. Feature data is standardised to zero mean and unit variance using parameters computed from the respective training set. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.music-ir.org/mirex/abstracts/2010/FE1.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
