PDF《CADENCE DETECTION IN WESTERN》

CADENCE DETECTION IN WESTERN TRADITIONAL STANZAIC SONGS USING MELODIC AND TEXTUAL FEATURES Peter van Kranenburg, Folgert Karsdorp Meertens Institute, Amsterdam, Netherlands {peter.

van.kranenburg,folgert.karsdorp}@meertens.knaw.nl ABSTRACT Many Western songs are hierarchically structured in stan- zas and phrases. The melody of the song is repeated for each stanza, while the lyrics vary. Each stanza is subdi- vided into phrases. It is to be expected that melodic and textual formulas at the end of the phrases offer intrinsic clues of closure to a listener or singer. In the current paper we aim at a method to detect such cadences in symbolically encoded folk songs. We take a trigram approach in which we classify trigrams of notes and pitches as cadential or as non-cadential. We use pitch, contour, rhythmic, textual, and contextual features, and a group of features based on the conditions of closure as stated by Narmour [11]. We employ a random forest classi?cation algorithm. The pre- cision of the classi?er is considerably improved by taking the class labels of adjacent trigrams into account. An abla- tion study shows that none of the kinds of features is suf?- cient to account for good classi?cation, while some of the groups perform moderately well on their own. 1. INTRODUCTION This paper presents both a method to detect cadences in Western folk-songs, particularly in folk songs from Dutch oral tradition, and a study to the importance of various mu- sical parameters for cadence detection. There are various reasons to focus speci?cally on ca- dence patterns. The concept of cadence has played a major role in the study of Western folk songs. In several of the most important folks song classi?cation systems, cadence tones are among the primary features that are used to put the melodies into a linear ordering. In one of the earli- est classi?cation systems, devised by Ilmari Krohn [10], melodies are ?rstly ordered according to the number of phrases, and secondly according to the sequence of ca- dence tones. This method was adapted for Hungarian mel- odies by B? artok and Kod? aly [16], and later on for German folk songs by Suppan and Stief [17] in their monumental Melodietypen des Deutschen Volksgesanges. Bronson [3] introduced a number of features for the study of Anglo- American folk song melodies, of which ?nal cadence and c Peter van Kranenburg, Folgert Karsdorp. Licensed under a Creative Commons Attribution 4.0 International Li- cense (CC BY 4.0). Attribution: Peter van Kranenburg, Folgert Kars- dorp. Cadence Detection in Western Traditional Stanzaic Songs using Melodic and Textual Features , 15th International Society for Music In- formation Retrieval Conference, 2014. mid-cadence are the most prominent ones. One of the underlying assumptions is that the sequence of cadence tones is relatively stable in the process of oral transmis- sion. Thus, variants of the same melody are expected to end up near to each other in the resulting ordering. From a cognitive point of view, the perception of clo- sure is of fundamental importance for a listener or singer to understand a melody. In terms of expectation [8, 11], a ?nal cadence implies no continuation at all. It is to be expected that speci?c features of the songs that are related to closure show different values for cadential patterns as compared to non-cadential patterns. We include a subset of features that are based on the conditions of closure as stated by Narmour [11, p.11]. Cadence detection is related to the problem of segmen- tation, which is relevant for Music Information Retrieval [21]. Most segmentation methods for symbolically repre- sented melodies are either based on pre-de?ned rules [4, 18] or on statistical learning [1,9,12]. In the current paper, we focus on the musical properties of cadence formulas rather than on the task of segmentation as such. Taking Dutch folk songs as case study, we investigate whether it is possible to derive a general model of the mel- odic patterns or formulas that speci?cally indicate melodic cadences using both melodic and textual features. To ad- dress this question, we take a computational approach by employing a random forest classi?er (Sections