The Role of Expectation and Probabilistic Learning in Auditory Boundary Perception: A Model Comparison

Abstract

Introduction Grouping and boundary perception are central to the understanding and modelling of core tasks in many areas of cognitive science. They are fundamental processes in, for example, natural language processing (eg speech segmentation and word discoveryöBrent 1999b; Jusczyk 1997), motor learning (eg identifying behavioural episodes öReynolds et al 2007; Newtson 1973), memory storage and retrieval (eg chunkingöKurby and Zacks 2007) and visual perception (eg analysing spatial organisation öMarr 1982). Our focus in this paper is on the perception and cognition of music (Krumhansl 1990; Temperley 2001), where the process by which the human perceptual system groups sequential musical elements together is one of the most fundamental issues. In particular, we examine the grouping of musical elements into contiguous segments that occur sequentially in time or, to put it another way, the identification of boundaries between the final element of one segment and the first element of the subsequent one. This way of structuring a musical surface is usually referred to as grouping (Lerdahl and Jackendoff 1983) or segmentation (Cambouropoulos 2006). We distinguish this kind of perceptual aggregation of auditory elements from the integration, or fusion, of auditory elements that occur simultaneously in time and also from the segregation of parallel auditory streams (Bregman 1990). In musical terms, the kinds of groups we consider correspond with motifs, phrases, sections and other aspects of musical form. We use the term grouping structure to refer to a piece of music structured in this way. It is taken that, just as speech is perceptually segmented into phonemes, and then words which subsequently provide the building blocks for the perception of phrases and complete utterances (Brent 1999b; Jusczyk 1997), motifs or phrases in music are identified by listeners, stored in memory and made available for inclusion in higher-level structural groups (Lerdahl and Jackendoff 1983; Peretz 1989; Tan et al 1981). The lowlevel organisation of the musical surface into groups allows the use of these primitive perceptual units in more complex structural processing and may alleviate demands on memory.

Symbolic Segmentation: A Corpus-Based Analysis of Melodic Phrases

“…That is, local leaps in duration make for a relatively strong predictor of phrase boundaries. This observation agrees with comparative studies [2,3,11,21]. We can not, however, generalize this finding to all music, as our study only investigated two musical traditions, and may contain biases introduced by the boundary annotation process used in the EFSC.…”

“…Recent comparative studies [2,3,11,12] report results of only modest success (F-scores [13] peaking at 0.60 − 0.66 when evaluated in large melodic corpora). Two models that have consistently ranked higher in these studies are LBDM [6] and Grouper [7].…”

“…Segmenting musical input is concerned with modelling the formation of temporal units holding musical content. Historically, the tasks of music segmentation that have received more attention within music content research fall into three groups: (a) the segmentation of musical audio into notes, as part of transcription systems [1], (b) the segmentation of symbolic encodings of music into phrases [2,3], and (c) the segmentation of audio/symbolic music files into sections [4]. In this paper we focus on the study on units of the second kind, i.e.…”

Multi-scale Modelling of Segmentation

“…Paulus, Müller, & Klapuri, 2010;Peeters & Deruty, 2009). Bruderer (2008), Wiering, de Nooijer, Volk, and Tabachneck-Schijf (2009), and Pearce et al (2010) have compared the performance of some segmentation systems. Other work on segmentation includes a neural study on finding working memory triggers (Burunat, Alluri, Toiviainen, Numminen, & Brattico, 2014) and a performance study on improvisational structure (Dean, Bailes, & Drummond, 2014).…”

Compression-based Modelling of Musical Similarity Perception

“…Narmour, 1990;Schellenberg, 1996;Schellenberg, 1997), suggesting that expectation reflects a process of statistical learning and probabilistic generation of predictions (Hansen & Pearce, 2014;Pearce, 2005;. IDyOM has also been used to predict perceived phrase endings at troughs in the information content profile (Pearce & Wiggins, 2006;Pearce, Müllensiefen, & Wiggins, 2010). The present work extends IDyOM to modelling perceived similarity between musical sequences using the compression distances defined above.…”