[show abstract][hide abstract] ABSTRACT: Typical signal-based approaches to extract musical de- scriptions from audio only have limited precision. A pos- sible explanation is that they do not exploit context, which provides important cues in human cognitive processing of music: e.g. electric guitar is unlikely in 1930s music, chil - dren choirs rarely perform heavy metal, etc. We propose an architecture to train a large set of binary classifiers si- multaneously, for many different musical metadata (genre, instrument, mood, etc.), in such a way that correlation be- tween metadata is used to reinforce each individual classi- fier. The system is iterative: it uses classification decisio ns it made on some classification problems as new features for new, harder problems; and hybrid: it uses a signal clas- sifier based on timbre similarity to bootstrap symbolic in- ference with decision trees. While further work is needed, the approach seems to outperform signal-only algorithms by 5% precision on average, and sometimes up to 15% for traditionally difficult problems such as cultural and sub- jective categories.
Proceedings of the 8th International Conference on Music Information Retrieval, ISMIR 2007, Vienna, Austria, September 23-27, 2007; 01/2007
[show abstract][hide abstract] ABSTRACT: The IST project Cuidado, which started in January 2001, aims at producing the first entirely automatic chain for extracting and exploiting musical metadata for browsing music. The Sony CSL laboratory is primarily interested in the context of popular music browsing in large-scale catalogues. First, we are interested in human-centred issues related to browsing "Popular Music". Popular here means that the music accessed to is widely distributed, and known to many listeners. Second, we consider "popular browsing" of music, i.e. making music accessible to non specialists (music lovers), and allowing sharing of musical tastes and information within communities, departing from the usual, single user view of digital libraries. This research project covers all areas of the music-to-listener chain, from music description - descriptor extraction from the music signal, or data mining techniques -, similarity based access and novel music retrieval methods such as automatic sequence generation, and user interface issues. This paper describes the scientific and technical issues at stake, and the results obtained, in the current state of the IST project.
Multimedia Tools and Applications 01/2006; 30:331-349. · 1.01 Impact Factor
[show abstract][hide abstract] ABSTRACT: We describe an incremental filtering algorithm to quickly compute the N nearest neighbors according to a similarity measure in a metric space. The algorithm exploits an in- trinsic property of a large class of similarity measures for which some parameter p has a positive influence both on the precision and the cpu cost (precision-cputime trade- off ). The algorithm uses successive approximations of the measure to compute first cheap distances on the whole set of possible items, then more and more expensive measures on smaller and smaller sets. We illustrate the algorithm on the case of a timbre similarity algorithm, which com- pares gaussian mixture models using a Monte Carlo ap- proximation of the Kullback-Leibler distance, where p is the number of points drawn from the distributions. We describe several Monte Carlo algorithmic variants, which improve the convergence speed of the approximation. On this problem, the algorithm performs more than 30 times faster than the naive approach.
ISMIR 2005, 6th International Conference on Music Information Retrieval, London, UK, 11-15 September 2005, Proceedings; 01/2005
FSDK'02, Proceedings of the 1st International Conference on Fuzzy Systems and Knowledge Discovery: Computational Intelligence for the E-Age, 2 Volumes, November 18-22, 2002, Orchid Country Club, Singapore; 01/2002
[show abstract][hide abstract] ABSTRACT: IBISA (Image-Based Identification/Search for Archaeology) is a research project supported by the French CNRS. The corresponding software tool manages databases of digital images of archaeological objects, and allows the user to perform searches by examples. For now, the system works with ancient (greek, roman) coins, and the generalization to medieval tiles is under progress. IBISA was designed to help the user decide, from their images, if two objects are either the same, come from the same matrix, share resemblance in style, or are completely different. It uses computer vision methods to make this decision while getting rid of the viewing conditions when searching for similarities in the databases. First, a segmentation method based on active contours extracts the useful part of each image from its background context. Then, a registration method based on the Fourier-Mellin transform sorts the images by similarity, canceling any translation, rotation, or zoom inherent to the photography.