Abstract | ||
---|---|---|
This paper applies the framework of robust principal components analysis (RPCA) to the problem of classifying acoustic soundscapes. RPCA provides a mechanism to decompose a data matrix as the sum of a low-rank matrix and a sparse matrix. In the context of data representing acoustic soundscapes, the low-rank matrix represents the slow-changing background sound events, and the sparse matrix represents the occasional foreground sound events. The data representations are obtained as feature embeddings from pretrained deep convolutional networks. The paper investigates the effectiveness of classifying acoustic soundscapes by using the foreground or background information alone. Further, by using the subspace projection technique of nuisance attribute projection (NAP), the undesired components from the foreground or background are removed. Our results indicate that RPCA and subspace projections in-deed provide benefits in improving discrimination for classifying acoustic soundscapes. |
Year | DOI | Venue |
---|---|---|
2020 | 10.23919/Eusipco47968.2020.9287875 | 2020 28th European Signal Processing Conference (EUSIPCO) |
Keywords | DocType | ISSN |
Acoustic scene classification,robust PCA,sub-space projections | Conference | 2219-5491 |
ISBN | Citations | PageRank |
978-1-7281-5001-7 | 0 | 0.34 |
References | Authors | |
5 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Dhanunjaya Varma Devalraju | 1 | 0 | 0.34 |
Padmanabhan Rajan | 2 | 22 | 7.63 |
A. D. Dileep | 3 | 15 | 7.72 |