We present the query-by-description (QBD) component of "Kandem," a time-aware music retrieval system. The QBD system we describe learns a relation between descriptive text concerning a musical artist and their actual acoustic output, making such queries as "Play me something loud with an electronic beat" possible by merely analyzing the audio content of a database. We show a novel machine learning technique based on Regularized Least-Squares Classification (RLSC) that can quickly and efficiently learn the non-linear relation between descriptive language and audio features by treating the problem as a large number of possible output classes linked to the same set of input features. We show how the RLSC training can easily eliminate irrelevant labels. I.

Abstract. As digital music collections grow, so does the need to organizing them automatically. In this paper we present an approach to hierarchically organize music collections at the artist level. Artists are grouped according to similarity which is computed using a web search engine and standard text retrieval techniques. The groups are described by words found on the webpages using term selection techniques and domain knowledge. We compare different term selection techniques, present a simple demonstration, and discuss our findings. 1

In this paper we present MusicRainbow which is a simple interface for discovering artists where colors encode different types of music. MusicRainbow is based on a new audiobased approach to compute artist similarity. This approach scores 15 percentage points higher in a genre classification task than the similarity computed on track level. Using a traveling salesman algorithm, similar artists are mapped near each other on a circular rainbow. Furthermore, we present a new approach of combining this audio-based information with information from the web. In particular, we label the rainbow and summarize the artists with words extracted from web pages related to the artists. We use different vocabularies for different hierarchical levels and heuristics to select the most descriptive labels. We conclude with a discussion of the results. The first impressions are very promising. 1.

Future MIR systems will be of great use and pleasure for potential users. If researchers have a clear picture about their “customers ” in mind they can aim at building and evaluating their systems exactly inside the different socio-cultural environments of such music listeners. Since music is in most cases embedded into a socio-cultural process we propose especially to evaluate MIR applications outside the lab during daily activities. For this purpose we designed a mobile music recommendation system relying on a trimodal music similarity metric, which allows for subjective on-the-fly adjustments of recommendations. It offers online access to large-scale metadata repositories as well as an audio database containing 1000 songs. We did first smallscale evaluations of this approach and came to interesting results regarding the perception of song similarity concerning the relations between sound, cultural issues and lyrics. Our paper will also give insights to the three different underlying approaches for song similarity computation (sound, cultural issues, lyrics), focusing in detail on a novel clustering of album reviews as found at online music retailers. Keywords: Socio-cultural issues in MIR, multimodal song similarity, ecological validation. 1.

Detecting artists that can be considered as prototypes for particular genres or styles of music is an interesting task. In this paper, we present an approach that ranks artists according to their prototypicality. To calculate such a ranking, we use asymmetric similarity matrices obtained via co-occurrence analysis of artist names on web pages. We demonstrate our approach on a data set containing 224 artists from 14 genres and evaluate the results using the rank correlation between the prototypicality ranking and a ranking obtained by page counts of search queries to Google that contain artist and genre. High positive rank correlations are achieved for nearly all genres of the data set. Furthermore, we elaborate a visualization method that illustrates similarities between artists using the prototypes of all genres as reference points. On the whole, we show how to create a prototypicality ranking and use it, together with a similarity matrix, to visualize a music repository.

We explore a simple, web-based method for predicting the genre of a given artist based on co-occurrence analysis, i.e. analyzing co-occurrences of artist and genre names on music-related web pages. To this end, we use the page counts provided by Google to estimate the relatedness of an arbitrary artist to each of a set of genres. We investigate four different query schemes for obtaining the page counts and two different probabilistic approaches for predicting the genre of a given artist. Evaluation is performed on two test collections, a large one with a quite general genre taxonomy and a quite small one with rather specific genres. Since our approach yields estimates for the relatedness of an artist to every genre of a given genre set, we can derive genre distributions which incorporate information about artists that cannot be assigned a single genre. This allows us to overcome the inflexible artist-genre assignment usually used in music information systems. We present a simple method to visualize such genre distributions with our Traveller’s Sound Player. Finally, we briefly outline how to adapt the presented approach to extract other properties of music artists from the web.

MusicSun is a graphical user interface to discover artists. Artists are recommended based on one or more artists selected by the user. The recommendations are computed by combining 3 different aspects of similarity. The users can change the impact of each of these aspects. In addition words are displayed which describe the artists selected by the user. The user can select one of these words to focus the search on a specific direction. In this paper we present the techniques used to compute the recommendations and the graphical user interface. Furthermore, we present the results of an evaluation with 33 users. We asked them, for example, to judge the usefulness of the different interface components and the quality of the recommendations. 1

We describe and discuss our recent work developing a database, methodology and ground...

The ever-growing amount of available music induces an increasing demand for Music Information Retrieval (MIR) applications such as music recommendation applications or automatic classification algorithms. When audio-based, a crucial part of such systems are the audio feature extraction routines. In this paper, we evaluate how well a variety of combinations of feature extraction and machine learning algorithms are suited to classify music into perceptual categories. The examined categorizations are perceived tempo, mood (happy / neutral /sad), emotion (soft / neutral / aggressive), complexity, and vocal content. The aim is to contribute to the investigation which aspects of music are not captured by the common audio descriptors; from our experiments we can conclude that most of the examined categorizations are not captured well. This indicates that more research is needed on alternative (possibly extra-musical) sources of information for useful music classification. 1.

Abstract. Today, among the best-performing audio-based music similarity measures are algorithms based on Mel Frequency Cepstrum Coefficients (MFCCs). In these algorithms, each music track is modelled as a Gaussian Mixture Model (GMM) of MFCCs. The similarity between two tracks is computed by comparing their GMMs. One drawback of this approach is that the distance space obtained this way has some undesirable properties. In this paper, a number of approaches to correct these undesirable properties are investigated. They use knowledge about the properties of music by using other music tracks as a reference. These reference tracks can either be the music collection itself, or they may be an external set of reference tracks. Our results show that the proposed techniques clearly improve the quality of this audio similarity measure. Furthermore, preliminary experiments indicate that the techniques also help to improve other similarity measures. They may even be useful in completely different domains, most notably text information retrieval. 1