Music Genre Classification using the multivariate AR feature integration model |
Peter Ahrendt, Anders Meng
|
| Abstract | Music genre classification systems are normally build as
a feature extraction module followed by a classifier. The
features are often short-time features with time frames of
10-30ms, although several characteristics of music require
larger time scales. Thus, larger time frames are needed to
take informative decisions about musical genre. For the
MIREX music genre contest several authors derive long
time features based either on statistical moments and/or
temporal structure in the short time features. In our contribution
we model a segment (1.2 s) of short time features
(texture) using a multivariate autoregressive model. Other
authors have applied simpler statistical models such as the
mean-variance model, which also has been included in
several of this years MIREX submissions, see e.g. Tzanetakis
(2005); Burred (2005); Bergstra et al. (2005); Lidy
and Rauber (2005). |
| Keywords | Multivariate Autoregressive Model, Music Genre Classification |
| Type | Misc [Other] |
| Journal/Book/Conference | MIREX 2005 (Contest on Music Genre Classification) |
| Year | 2005 Month August |
| Note | Extended Abstract |
| Electronic version(s) | [pdf] |
| BibTeX data | [bibtex] |
| IMM Group(s) | Intelligent Signal Processing |