Shaping Structure Through Sound: Effects of Repeated Listening on Musical Segmentation in Musicians and Non-Musicians

Shaping Structure Through Sound: Effects of Repeated Listening on Musical Segmentation in Musicians and Non-Musicians Duri Bae1,Ahyeon Choi2and Kyogu Lee2,∗ 1School of Biological Science, Seoul National University 2Department of Intelligence and Information, Seoul National University Abstract Repeated listening gradually shapes listeners’ perception of musical structure, making it a critical behavioral factor to consider in music information retrieval (MIR) systems. However, how this process varies with musical training remains unclear. This study investigates how musicians and non-musicians segment music over repeated exposure to an unfamiliar piece. Thirty-three participants (13 music majors, 20 non-majors) listened five times to the second movement of Bartók’s Piano Sonata Sz. 80, marking perceived structural boundaries during each trial. Their responses were compared to two analytical models: Traditional Harmonic Analysis (TA) and Schenkerian Analysis (SA). Over time, music majors increasingly aligned with TA, while non-majors showed greater variability without converging on a consistent framework. Statistical analysis confirmed a significant group difference in response consistency during later trials. These findings suggest that musical training facilitates reliance on structural schemata, whereas those without formal training rely more on perceptual features such as dynamics or mood. By revealing group-specific strategies in structural listening, this study contributes to a human-centered understanding of musical segmentation and offers empirical insights for designing adaptive music information retrieval systems that account for listener expertise and perceptual tendencies. Keywords repeated listening, musical segmentation, music structure 1. Introduction One of the fundamental characteristics of music is its temporality; a musical piece cannot be fully perceived until it is experienced from beginning to end. Listening involves more than detecting vibrations—it requires cognitive processing that interprets and assigns meaning to auditory stimuli, making music perception a complex, multi-layered experience. Repeated listening, in particular, helps listeners uncover structural details that contribute to a deeper understanding of musical form [ 1 ]. Consequently, repeated exposure is considered essential for grasping a piece’s formal structure. Beyond its cognitive significance, repeated listening is also relevant to human-centered Music Information Retrieval (MIR). MIR is a multidisciplinary field that combines signal processing, information science, and cognitive research to analyze and retrieve music content effectively [ 2 ]. Understanding how listeners perceive and segment music over time can inform adaptive MIR systems that reflect individual perception and musical expertise. Previous studies suggest that listeners become increasingly familiar with musical material through repetition and tend to organize its elements in a more conceptual manner [ 1 ]. However, schematic expectations for conventional patterns like tonic chords may be weaker than factual expectations formed by repetition [ 3 ]. The Generative Theory of Tonal Music (GTTM), for instance, describes how listeners derive hierarchical structure through perceptual and cognitive rules such as grouping and metrical structure [4], framing music as a type of grammar internalized over time. Despite growing theoretical interest in these cognitive processes, empirical evidence on how repeated listening shapes structural perception remains limited. Without such data, theoretical accounts remain HCMIR25: 3rd Workshop on Human-Centric Music Information Research, September 20, 2025, Daejeon, Korea ∗Corresponding author. Envelope-Open[email protected] (D. Bae); [email protected] (A. Choi); [email protected] (K. Lee) © 2025 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). speculative. There is thus a need for empirical studies investigating how listeners with varying levels of expertise perceive musical form through repeated exposure. In this study, we define “musical structure” as the macrostructural organization of a piece, typically described by cadences, phrases, rhythms, and harmonic progressions [ 5 ]. While early harmonic analysis focused on surface-level chord progressions [ 6 ], Heinrich Schenker proposed a hierarchical approach emphasizing underlying tonal coherence [7,8]. To evaluate how these analytical frameworks relate to actual listening behavior, we examine how repeated listening affects structural perception in listeners with and without formal musical training. As a stimulus, we selected the second movement of Bartók’s Piano Sonata, Sz. 80 (1926), which features ambiguous formal boundaries and has been interpreted both through traditional harmonic analysis (TA) and Schenkerian analysis (SA) [9,10,11]. This study asks: Does repeated listening guide listeners toward alignment with formal structural models? If so, which model—TA or SA—does it resemble more closely? Are there consistent differences between music majors and non-majors in how structure is perceived? By addressing these questions, we aim to clarify how expertise and exposure interact in the perception of musical form and offer insight into perceptual modeling for listener-adaptive MIR systems. 2. Methods 2.1. Participants A total of 36 participants (15 music majors and 21 non-music majors) took part in the study. Music majors were defined as individuals who had received at least five years of formal music education. Three participants (one from each group and one for personal reasons) were excluded due to prior familiarity with the piece or other factors, resulting in 33 participants (13 music majors and 20 non-majors) in the final analysis. All music majors were familiar with key theoretical concepts such as sonata form, ternary structure, and harmony through both theoretical and practical training. All participants provided written informed consent and received financial compensation for their participation. Ethics approval was obtained by the Institutional Review Board. 2.2. Stimuli The second movement of Bartók’s Piano Sonata, Sz. 80 (1926) was used as the stimulus. Known for its ambiguous formal boundaries and dissonant harmonic language, the piece has been analyzed using both Schenkerian Analysis (SA) and Traditional Harmonic Analysis (TA) [ 8 , 9 ]. According to SA, segmentation points were identified at mm14 (73s) and mm47 (207s). TA segmented the movement into an A–B–A’ ternary form, with boundaries at mm29 (134) and mm42 (183s). Additionally, TA considers mm14 (73s) as the boundary between Part 1 and Part 2 within section A (see Table 2). 2.3. Procedure Participants completed a demographic questionnaire and were informed of the piece’s three-part structure. The experiment, implemented with the pygame library in Python, involved five repeated listenings during which participants marked perceived boundaries using the space bar or a stopwatch. measure(mm) 1 14 29 42 47 62 SA A B C TA A1A2B A′ 1 Time(sec) 0 73 134 183 207 272 Table 1 The segmentation of the musical form as analyzed by SA and TA. SA categorizes the piece into three distinct sections, whereas TA classifies it into an A–B–A’ ternary form. Figure 1: Experiment Process. A black screen was displayed during playback, and note-taking was permitted. A post-experiment questionnaire assessed familiarity with the piece and knowledge of related musical forms. Figure 1 summarizes the procedure. 2.4. Analysis Segmentation distributions were compared across groups and trials using Python-based violin plots, standard deviations, and Levene’s test to evaluate intra-group consistency and group variance. 3. Results Figure 2illustrates segmentation responses across trials. At the first segmentation point, both groups tended to respond near mm14 (73s, SA), with no significant group differences. However, non-majors increasingly deviated from the TA-defined boundary (mm29; 134s), while music majors showed a modest alignment with TA by the fifth trial when outliers (>mm58; >250s) were excluded. Figure 2: Segmentation Points by Trials and Major. Violin plot showing segmentation responses across trials. Music majors (light gray) increasingly aligned with TA boundaries (mm29, mm42; dashed), while non-majors (black) remained dispersed. SA boundaries are dotted-dashed (mm14, mm47). In contrast, the second segmentation point revealed a clear divergence between groups. While both groups were initially scattered, music majors progressively converged toward the TA boundary at mm42 from the fourth trial onward. This pattern persisted in the fifth trial. Non-majors continued to show broad dispersion, suggesting no consistent structural reference point. Standard deviation trends (Fig. 3) support these findings. Music majors showed consistent reductions in variability, especially for segmentation point 2. In the fifth trial, their standard deviation was lowest overall. Levene’s test confirmed no group difference at segmentation point 1, but a significant difference for point 2 in the final trial (𝑝 = .03), indicating stronger internal agreement among majors (Table 2). Figure 3: Standard Deviation of Segmentation Points. Solid lines, square = point 2; dashed, circle = point 1; light gray = majors; black = non-majors. Trials 1 2 3 4 5 Seg. 1 0.29 0.33 0.92 0.49 0.15 Seg. 2 0.88 0.96 0.26 0.31 0.03* Table 2 Levene’s Test P-values (Major vs Non-major). *𝑝 < .05 for final trial, segmentation point 2. These results reveal differential effects of repeated listening across structural positions. At segmentation point 1, both groups favored the SA boundary (mm14; 73s), possibly due to local surface cues such as phrase closure or texture change. The lack of convergence on the TA boundary (mm29; 134s) suggests that it may be less salient or perceptually distinct. At point 2, however, TA’s mm42 (183s) boundary appears to be more cognitively accessible, particularly to trained listeners, likely due to its alignment with recurring harmonic and formal cues. Music majors’ convergence suggests internalization of high-level structure through exposure. Their notes frequently referenced harmonic progression or motivic recurrence. In contrast, non-majors remained scattered and referenced “mood changes” or instrument density shifts. This aligns with research showing that non-experts segment music based on perceptual salience rather than abstract structure [ 12 , 13 , 14 ]. Levene’s test confirms this: only at the second segmentation point and only among trained listeners did variance reliably decrease, indicating schema-based alignment. This pattern mirrors findings from ERP studies, where closure-related responses (CPS) are more robust and hierarchically distributed in trained listeners [15]. 4. Discussion This study shows that repeated listening facilitates structural alignment primarily in listeners with formal training. Music majors progressively converged on TA-defined boundaries, particularly at the second segmentation point, suggesting that repeated exposure supports integration of harmonic and formal cues. In contrast, non-majors exhibited greater variability, likely reflecting reliance on perceptual or subjective features such as dynamics or mood. These results suggest that segmentation is shaped not only by acoustic features but also by internalized analytical schemas developed through training. The contrasting patterns between segmentation points 1 and 2 further support the idea that some boundaries are more perceptually salient than others. Nonetheless, this study has limitations. The use of a single musical piece and the restriction to three segmentation points may limit generalizability and reduce participants’ ability to express intuitive structure. Future studies could employ a broader range of works and more flexible response formats, while neuroimaging tools such as fMRI [ 16 ] may further reveal neural mechanisms underlying expert versus novice strategies. These findings also inform MIR applications by highlighting listener-dependent segmentation, which could guide adaptive systems tailored to user expertise and listening context. References [1] L. Pollard-Gott, “Emergence of thematic concepts in repeated listening to music,” Cognitive Psychology, vol. 15, no. 1, pp. 66–94, 1983. [2] J. S. Downie, “Music information retrieval,” Annual Review of Information Science and Technology, vol. 37, no. 1, pp. 295–340, 2003. [3] B. Tillmann and E. Bigand, “Musical structure processing after repeated listening: Schematic expectations resist veridical expectations,” Musicae Scientiae, vol. 14, no. 2_suppl, pp. 33–47, 2010. [4] F. Lerdahl and R. Jackendoff, A Generative Theory of Tonal Music. Cambridge, MA: MIT Press, 1983. [5] A. K. McNamee, Tonal Harmony, with an Introduction to Twentieth-Century Music, 1986. [6] T. Christensen, Rameau and Musical Thought in the Enlightenment, vol. 4, Cambridge University Press, 2004. [7] A. Forte, “Schenker’s conception of musical structure,” Journal of Music Theory, vol. 3, no. 1, pp. 1–30, 1959. [8] K. Agawu, The Art of Tonal Analysis: Twelve Lessons in Schenkerian Theory, 2018. [9] Y. L. Chang, “A study and performer’s analysis of piano sonatas by Béla Bartók and Serge Prokofiev,” Ph.D. dissertation, University of Cincinnati, 2000. [10] J. Lee, “A linear analysis of piano sonata (1926) Sz. 80 by Bela Bartok: The genesis and development of the composition,” Ph.D. dissertation, University of North, 2023. [11] H.-J. Choi, “A Study on Béla Bartók’s <Piano Sonata, Sz. 80>,” Master’s thesis, Pyongtaek University, 2015. [12] J. Lee, T. Hill, and L. Work, “What does music mood mean for real users?” in Proceedings of the 2012 iConference, pp. 112–119, 2012. [13] L. L. Cuddy, A. J. Cohen, and D. J. K. Mewhort, “Perception of structure in short melodic sequences,” Journal of Experimental Psychology: Human Perception and Performance, vol. 7, no. 4, pp. 869–883, 1981. [14] M. Phillips, C. Collins, and M. D. Heron, “What determines the perception of segmentation in contemporary music?” Frontiers in Psychology, vol. 11, Article 1001, 2020. [15] J. Zhang, C. Jiang, L. Zhou, and Y. Yang, “Perception of hierarchical boundaries in music and its modulation by expertise,” Neuropsychologia, vol. 91, pp. 490–498, 2016. [16] C. Toader, C. P. Tataru, I.-A. Florian, R.-A. Covache-Busuioc, B.-G. Bratu, L. A. Glavan, A. Bordeianu, D.-I. Dumitrascu, and A. V. Ciurea, “Cognitive crescendo: how music shapes the brain’s structure and function,” Brain Sciences, vol. 13, no. 10, p. 1390, 2023.