On similarity earch in audio signals using adaptive sparse approximations

We explore similarity search in data compressed and described by adaptive methods of sparse approximation, specifically audio signals. The novelty of this approach is that one circumvents the need to compute and store a database of features since sparse approximation can simultaneously provide a description and compression of data. We investigate extensions to a method previously proposed for similarity search in a homogenous image database using sparse approximation, but which has limited applicability to search heterogeneous databases with variable-length queries - necessary for any useful audio signal search procedure. We provide a simple example as a proof of concept, and show that similarity search within adapted sparse domains can provide fast and efficient ways to search for data similar to a given query.