Nonlinear dimensionality reduction for efficient and effective audio similarity searching

Authors:
Dimitris Rafailidis;Alexandros Nanopoulos;Yannis Manolopoulos
Affiliations:
Department of Informatics, Aristotle University, Thessaloniki, Greece 54124;Institute of Computer Science, University of Hildesheim, Hildesheim, Germany 31141;Department of Informatics, Aristotle University, Thessaloniki, Greece 54124
Venue:
Multimedia Tools and Applications
Year:
2011

Citing 8
Cited 2

Locally adaptive dimensionality reduction for indexing large time series databases

ACM Transactions on Database Systems (TODS)
R-trees: a dynamic index structure for spatial searching

SIGMOD '84 Proceedings of the 1984 ACM SIGMOD international conference on Management of data
M-tree: An Efficient Access Method for Similarity Search in Metric Spaces

VLDB '97 Proceedings of the 23rd International Conference on Very Large Data Bases
Similarity Search in High Dimensions via Hashing

VLDB '99 Proceedings of the 25th International Conference on Very Large Data Bases
A Large-Scale Evaluation of Acoustic and Subjective Music-Similarity Measures

Computer Music Journal
A non-linear dimensionality-reduction technique for fast similarity search in large databases

Proceedings of the 2006 ACM SIGMOD international conference on Management of data
The use of the area under the ROC curve in the evaluation of machine learning algorithms

Pattern Recognition
Rapid and brief communication: Incremental locally linear embedding

Pattern Recognition

A fast audio similarity retrieval method for millions of music tracks

Multimedia Tools and Applications
A unified framework for multimodal retrieval

Pattern Recognition

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, we address the issue of nonlinear dimensionality reduction to efficiently index spectral audio similarity measures. We propose the embedding of the spectral similarity space to a low-dimensional Euclidean space. This guarantees the triangular inequality and allows the adoption of several indexing schemes. We enlighten the advantages of the proposed indexable method against recently proposed spectral similarity measures that are also indexable. Moreover, our method compares favorably to linear dimensionality reduction methods, like multidimensional scaling (MDS). The proposed method significantly reduces the computation time during the construction process compared to any audio measure and, simultaneously, minimizes the searching cost for similar songs. To the best of our knowledge, the important issue of audio similarity measures' scalability is addressed for the first time.