Share Email Print
cover

Proceedings Paper

On application of image analysis and natural language processing for music search
Author(s): Grzegorz Gwardys
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

In this paper, I investigate a problem of finding most similar music tracks using, popular in Natural Language Processing, techniques like: TF-IDF and LDA. I de ned document as music track. Each music track is transformed to spectrogram, thanks that, I can use well known techniques to get words from images. I used SURF operation to detect characteristic points and novel approach for their description. The standard kmeans was used for clusterization. Clusterization is here identical with dictionary making, so after that I can transform spectrograms to text documents and perform TF-IDF and LDA. At the final, I can make a query in an obtained vector space. The research was done on 16 music tracks for training and 336 for testing, that are splitted in four categories: Hiphop, Jazz, Metal and Pop. Although used technique is completely unsupervised, results are satisfactory and encouraging to further research.

Paper Details

Date Published: 25 October 2013
PDF: 7 pages
Proc. SPIE 8903, Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2013, 89030Z (25 October 2013); doi: 10.1117/12.2035450
Show Author Affiliations
Grzegorz Gwardys, Warsaw Univ. of Technology (Poland)


Published in SPIE Proceedings Vol. 8903:
Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2013
Ryszard S. Romaniuk, Editor(s)

© SPIE. Terms of Use
Back to Top