Anja Volk

A Comparison between Global and Local Features for Computational Classification of Folk Song Melodies

Abstract In computational approaches to the study of variation among folk song melodies from oral culture, both global and local features of melodies have been used. From a computational point of view, the representation of a melody as a vector of global feature values, each summarizing an aspect of the entire melody, is attractive. However, from an annotation study on perceived melodic similarity and human categorization in music it followed that local features of melodies are most important to classify and recognize melodies. We compare both approaches in a computational classification task. In both cases, the discriminative power of features is assessed. We use a feature evaluation criterion that is based on the performance of a nearest-neighbour classifier. As distance measure for vectors of global features, we use the Euclidian distance. For the sequences of local features, we use the score of the Needleman–Wunsch alignment algorithm. In each of our comparisons, the local features correspond to the global features. In all cases, it appears that the local approach outperforms the global approach in a classification task for melodies, which indicates that local features carry more information about the identity of melodies. Therefore, locality is a crucial factor in modelling melodic similarity among folk song melodies.

Melodic similarity among folk songs: An annotation study on similarity-based categorization in music:

In this article we determine the role of different musical features for the human categorization of folk songs into tune families in a large collection of Dutch folk songs. Through an annotation study we investigate the relation between musical features, perceived similarity and human categorization in music. We introduce a newly developed annotation method which is used to create an annotation data set for 360 folk song melodies in 26 tune families. This dataset delivers valuable information on the contribution of musical features to the process of categorization which is based on assessing the similarity between melodies. The analysis of the annotation data set reveals that the importance of single musical features for assessing similarity varies both between and within tune families. In general, the recurrence of short characteristic motifs is most relevant for the perception of similarity between songs belonging to the same tune family. Global melodic features often used for the description of melodies (such as melodic contour) play a less important role. The annotation data set is a valuable resource for further research on melodic similarity and can be used as enriched “ground truth” to test various kinds of retrieval algorithms in Music Information Retrieval. Our annotation study exemplifies that assessing similarity is crucial for human categorization processes, which has been questioned within Cognitive Science in the context of rule-based approaches to categorization.

Cognition-based Segmentation for Music Information Retrieval Systems

Abstract This paper investigates the generic problem of model selection in the specific context of Music Information Retrieval (MIR). In MIR research, similarity measures are developed for ranking musical items with respect to their relevance to a users musical query. The application of such similarity measures in MIR systems typically requires musical works to be divided into more manageable units. This involves two tasks: melody segmentation and voice separation. For both of these tasks, several computational models have been proposed in the symbolic domain. It seems reasonable to assume that those solutions that are most in accordance with human performance will result in the best ranking of retrieval output. We conducted two experiments, each with twenty experts and twenty novices. In the melody segmentation experiment, we found a high agreement between the participants. Evaluating algorithm output against participant data, we conclude that human output cannot be distinguished from three of the segmentation algorithms (Grouper, IDyOM and LBDM). For voice separation—which we evaluated by means of a melody identification task—the situation is different, as the combined results of two algorithms (Skyline and SSA) were shown to agree best with experimental results, and differences were found between novice and expert performance. Several other model selection criteria besides performance are discussed in conclusion.

The Study of Syncopation Using Inner Metric Analysis: Linking Theoretical and Experimental Analysis of Metre in Music

Abstract This paper investigates the influence of syncopation on the metric structure of musical pieces using the computational model of Inner Metric Analysis. Inner Metric Analysis generates metric hierarchies evoked by the note onsets of a piece. Syncopation in the rhythmic structure influences these hierarchies in different ways. The comparison of local and global perspectives on the metric structure allows one to distinguish between different amounts of syncopation present in a musical piece. This paper shows that the study of syncopation using Inner Metric Analysis contributes to explaining tapping performances of listeners. Hence comparing the structural descriptions generated by the model to results of a listening experiment helps to link music theoretic and perceptual studies.

Persistence and change: Local and global components of metre induction using Inner Metric Analysis

This paper compares local and global perspectives on the metric structure of musical compositions using the computational model of Inner Metric Analysis. The comparison addresses different hierarchical levels of compositions with respect to metric stability and change. Inner Metric Analysis generates metric hierarchies that are persistent from the global perspective and sensitive to changes from the local perspective. While the global perspective indicates the predominant metric characteristic of a piece or fragment, the local perspective distinguishes between regions of different metric characteristics. Like studies of harmony that investigate tonality as a hierarchic web of relations between chords, phrases and large sections, this article suggests that the time organization of a piece consists of different hierarchical levels. The description of local and global metric characteristics of Weberns Op. 27 using Inner Metric Analysis complements and extends arguments given by David Lewin about a metrical ...This paper compares local and global perspectives on the metric structure of musical compositions using the computational model of Inner Metric Analysis. The comparison addresses different hierarchical levels of compositions with respect to metric stability and change. Inner Metric Analysis generates metric hierarchies that are persistent from the global perspective and sensitive to changes from the local perspective. While the global perspective indicates the predominant metric characteristic of a piece or fragment, the local perspective distinguishes between regions of different metric characteristics. Like studies of harmony that investigate tonality as a hierarchic web of relations between chords, phrases and large sections, this article suggests that the time organization of a piece consists of different hierarchical levels. The description of local and global metric characteristics of Weberns Op. 27 using Inner Metric Analysis complements and extends arguments given by David Lewin about a metrical problem observed in this piece.

Modelling Folksong Melodies

Abstract In the second half of the twentieth century, ethnomusicologists assembled a collection of more than 7000 field recordings of Dutch ballads. Collectively known as Onder de groene linde, these recordings are preserved at the Meertens Institute in Amsterdam. Because of its size, composition and quality of metadata, Onder de groene linde is a unique resource for studying the musical properties of folksongs. For such study, it would be essential to search and order the songs automatically, not just using the metadata, but especially their musical content. It is the aim of the WITCHCRAFT project to design and implement methods for processing the musical content of the songs. Such a project involves two disciplines, musicology and computer science, that have different goals and methodologies. Such differences can lead to unproductive tensions, but can also be exploited in order to attain new insights that could not have been attained by the separate disciplines.

Finding Repeated Patterns in Music: State of Knowledge, Challenges, Perspectives

This paper discusses the current state of knowledge on musical pattern finding. Various studies propose computational methods to find repeated musical patterns. Our detailed review of these studies reveals important challenges in musical pattern finding research: different methods have not yet been directly compared, and the influence of music representation and filtering on the results has not been assessed. Moreover, we need a thorough understanding of musical patterns as perceived by human listeners. A sound evaluation methodology is still lacking. Consequently, we suggest perspectives for musical pattern finding: future research can provide a comparison of different methods, and an assessment of different music representations and filtering criteria. A combination of quantitative and qualitative methods can overcome the lacking evaluation methodology. Musical patterns identified by human listeners form a reference, but also an object of study, as computational methods can help us understand the criteria underlying human notions of musical repetition.

On Pitch and Chord Stability in Folk Song Variation Retrieval

In this paper we develop methods for computer aided folk song variation research. We examine notions and examples of stability for pitches and implied chords for a group of melodic variants. To do this we employ metrical accent levels, simple alignment techniques and visualization techniques. We explore how one can use insight into stability of a known set of variants to query for additional variants.

Mathematical and computational approaches to music: challenges in an interdisciplinary enterprise

This Special Issue is dedicated to explicate and discuss methodological issues in the interdisciplinary research field of mathematical and computational approaches to music. It arose from a lively panel discussion at the third International Conference on Mathematics and Computation in Music 2011 in Paris. We have organized this panel in order to initiate the much needed interdisciplinary dialogue on the How, Why, and What of our modelling of and theorizing about music in the wide field of science, humanities and cognitive approaches to music research. From the contributions of the three panelists to this Special Issue, we extract key topics that the interdisciplinary scientific community needs to address in order to enable the different disciplines to productively complement one another in achieving a comprehensive approach to music as a complex, yet fundamental human trait.

Comparing Computational Approaches to Rhythmic and Melodic Similarity in Folksong Research

In this paper we compare computational approaches to rhythmic and melodic similarity in order to find relevant features characterizing similarity in a large collection of Dutch folksongs. Similarity rankings based on Transportation Distances are compared to an approach of rhythmic similarity based on Inner Metric Analysis proposed in this paper. The comparison between the two models demonstrates the important impact of rhythmic organization on melodic similarity.