On Automatic Music Genre Recognition by Sparse Representation Classification using Auditory Temporal Modulations

A recent system combining sparse representation classification (SRC)
and a perceptually-based acoustic feature (ATM)
\cite{Panagakis2009,Panagakis2009b,Panagakis2010c},
outperforms by a significant margin the state of the art in music genre recognition, e.g., \cite{Bergstra2006}.
With genre so difficult to define,
and seemingly based on factors more broad than acoustics,
this remarkable result motivates investigation into, among other things,
why it works and what it means for how humans organize music.
In this paper, we review the application of SRC and ATM to recognizing genre,
and attempt to reproduce the results of \cite{Panagakis2009}.
First, we find that classification results
are consistent for features extracted from different analyses.
Second, we find that SRC accuracy improves
when we pose the sparse representation problem
with inequality constraints.
Finally, we find that only when we reduce the number of classes by half
do we see the high accuracies reported in \cite{Panagakis2009}.