Studying the ways to recommend music to a user is a central task within the music information research community. From a content-based point of view, this task can be regarded as obtaining a suitable distance measurement between songs defined on a certain feature space. We propose two such distance measures. First, a low-level measure based on tempo-related aspects, and second, a highlevel semantic measure based on regression by support vector machines of different groups of musical dimensions such as genre and culture, moods and instruments, or rhythm and tempo. We evaluate these distance measures against a number of state-of-the-art measures objectively, based on 17 ground truth musical collections, and subjectively, based on 12 listeners ’ ratings. Results show that, in spite of being conceptually different, the proposed methods achieve comparable or even higher performance than the considered baseline approaches. Furthermore, they open up the possibility to explore distance metrics that are based on truly semantic notions. 1.

Automatic music classification is a wide-ranging and multidisciplinary area of inquiry that offers significant benefits from both academic and commercial perspectives. This dissertation focuses on the development of jMIR, a suite of powerful, flexible, accessible and original software tools that can be used to design, share and apply a wide range of automatic music classification technologies. jMIR permits users to extract meaningful information from audio recordings, symbolic musical representations and cultural information available on the Internet; to use machine learning technologies to automatically build classification models; to automatically collect profiling statistics and detect metadata errors in musical collections; to perform experiments on large, stylistically diverse and well-labelled collections of music in both audio and symbolic formats; and to store and distribute information that is essential to automatic music classification in expressive and flexible standardised file formats. In order to have as diverse a range of applications as possible, care was taken to avoid tying jMIR to any particular types of music classification. Rather, it is designed to be a

The size of personal music collections has constantly increased over the past years. As a result, the traditional metadata based lists to browse these collections have reached their limits. Interfaces that are based on music similarity offer an alternative and thus are increasingly gaining attention. Music similarity is typically either derived from audiofeatures (objective approach) or from user driven information sources, such as collaborative filtering or social tags (subjective approach). Studies show that the latter techniques outperform audio-based approaches when it comes to describe the perceived music similarity. However, subjective approaches typically only define pairwise relations as opposed to the global notion of similarity given by audiofeature spaces. Many of the proposed interfaces for similarity based music access inherently depend on this global notion and are thus not applicable to user driven music similarity measures. The first contribution of this paper is a high dimensional music space that is based on user driven similarity measures. It combines the advantages of audiofeature spaces (global view) with the advantages of subjective sources that better reflect the users ’ perception. The proposed space compactly represents similarity and therefore is well suited for offline use, such as in mobile applications. To demonstrate the practical applicability, the second contribution is a comprehensive mobile music player that incorporates several smart interfaces to access the user’s music collection. Based on this application, we finally present a large-scale user study that underlines the benefits of the introduced interfaces and shows their great user acceptance.

This paper presents the jWebMiner 2.0 cultural feature extraction software and describes the results of several musical genre classification experiments performed with it. jWebMiner 2.0 is an easy-to-use and open-source tool that allows users to mine the Internet in order to extract features based on both Last.fm social tags and general web search string co-occurrences extracted using the Yahoo! API. The experiments performed found that the features based on social tags were more effective at classifying music into a small (5-genre) genre ontology, but the features based on general web co-occurrences were more effective at classifying a moderate (10-genre) ontology. It was also found that combining the two types of features resulted in improved performance overall. 1.

We present an overview of digital audio analysis and synthesis methods for sound design and composition. The sonic landscape available to us contains a multitude of sounds, ranging from artificial to natural, purely musical to purely “real-world. ” To take full advantage of this diversity, it is helpful to have a comprehensive knowledge of the tools with which we can create and manipulate different types of sounds. We offer a summary of existing techniques organized by their underlying technology and source material, as well as by the kinds of sounds for which they are known to be effective. Our survey aims to support the synthesis of rich sound scenes and environments by facilitating the selection of the most appropriate tools for each component sound. A full toolbox means the whole world need not look like a nail! 1.

This paper introduces a new method for improving the accuracy in medium scale music similarity problems. Recently, it has been shown that the raw accuracy of query by example systems can be enhanced by considering priors about the distribution of its output or the structure of the music collection being considered. The proposed approach focuses on reducing the dependency to those priors by considering an eigenvalue decomposition of the aforementioned system’s output. Experiments carried out in the framework of cover song detection show that the proposed approach has good performance for enhancing a high accuracy system. Furthermore, it maintains the accuracy level for lower performing systems. 1.

With the continuing advances in data storage and communication technology, there has been an explosive growth of music information from different application domains. As an effective technique for organizing, browsing, and searching large data collections, music information retrieval is attracting more and more attention. How to measure and model the similarity between different music items is one of the most fundamental yet challenging research problems. In this paper, we introduce a novel framework based on a multimodal and adaptive similarity measure for various applications. Distinguished from previous approaches, our system can effectively combine music properties from different aspects into a compact signature via supervised learning. In addition, an incremental Locality Sensitive Hashing algorithm has been developed to support efficient retrieval processes with different kinds of queries. Experimental results based on two large music collections reveal various advantages of the proposed framework including effectiveness, efficiency, adaptiveness, and scalability. Categories and Subject Descriptors

DeviceCycle: rapid and reusable prototyping of gestural interfaces, applied to audio browsing by similarity

representation of the spectral envelope of a given signal. Although they have been shown to be a powerful descriptor for speech and music signals, more accurate and easily interpretable options can be devised. In this study, we present and evaluate the shape-based spectral contrast descriptor, which is build up from the previously proposed octave-based spectral contrast descriptor. We compare the three aforementioned descriptors with regard to their discriminative power and MP3 compression robustness. Discriminative power is evaluated within a prototypical genre classification task. MP3 compression robustness is measured by determining the descriptor values ’ change between different encodings. We show that the proposed shape-based spectral contrast descriptor yields a significant increase in accuracy, robustness, and applicability over the octave-based spectral contrast descriptor. Our results also corroborate initial findings regarding the accuracy improvement of the octave-based spectral contrast descriptor over Mel-frequency cepstral coefficients for the genre classification task. 1

ISMS TR 2009-11000 This paper presents an extensive analysis of a sample of a social network of musicians. The network sample is first analyzed using standard complex network techniques to verify that it has similar properties to other web-derived complex networks. Content-based pairwise dissimilarity values between the musical data associated with the network sample are computed, and the relationship between those content-based distances and distances from network theory explored. Following this exploration, hybrid graphs and distance measures are constructed, and used to examine the community structure of the artist network. Finally, results of these investigations are presented and considered in the light of recommendation and discovery applications with these hybrid measures as their basis.