Humans can identify a specific melody within homophonic and polyphonic music because it often appears in a higher pitch range, making it easily distinguishable (Fujioka et al., 2005; Trainor et al., 2014). This phenomenon aligns with the figure–ground concept in Gestalt psychology (Köhler, 1967; Wagemans et al., 2012). Similar to a visual stimulus (Supplementary Figure 1), the melody can serve as the “figure,” while other voices constitute the “ground.” Additionally, listeners can sometimes shift their attention during a phrase, perceiving the melody as the background while other voices being dominant instead (Ragert et al., 2014; Deutch, 2019). However, even when the figure–ground relationship favors the figure as the more perceptually dominant voice, the ground can remain perceptible (Bigand et al., 2000), and vice versa.

Musical structure, including pitch, tonality, and harmony, is learned through experience, understanding of musical structure facilitates the recognition and anticipation of patterns in familiar pieces (Narmour, 2000; Tillmann et al., 2000). When a familiar melody appears in a musical piece, it is easily recognized the figure in the musical structure. However, repeated exposure to the melody may alter the figure–ground relationship between the upper and lower voices (Taher et al., 2016), and this change may eventually lead to the natural collapse of the figure–ground relationship centered on the upper voice of the familiar melody.

In our previous study using Mozart's 12 Variations, K. 265 (Kim et al., 2020), we observed that only frontotemporal connectivity between the left Heschl's gyrus (HG) and left inferior frontal gyrus (IFG) changed in response to the presence or absence of the “Twinkle, Twinkle, Little Star” (TTLS) melody of “C5-C5-G5-G5-A5.” This connectivity pattern for the TTLS was observed across a target phrase (T) of 2.1 s at the beginning of each variation. However, if the figure–ground relationship shifts after repetitions, the connectivity strength can become inconsistent (Supplementary Figure 1C). The present study examined how frontotemporal connectivity for Variations II and IV (Figure 1) changes across four target phrases (T1–T4) featuring the TTLS melody. We hypothesized that: (1) If the connectivity pattern does not differ significantly between Variations II and IV, the TTLS melody remains the figure, with the lower voice serving as the ground; and (2) If the connectivity pattern differs significantly after repetitions, the TTLS melody may not be the sole figure, as the lower voices influence its prominence.

A difference in frontotemporal connectivity from the left IFG to the right HG between the two variations was only observed in the final target phrase of T4 and not in the preceding three phrases (Figure 2A). As we hypothesized, frontotemporal connectivity showed inconsistency in the figure–ground relationship between the two variations of the TTLS melody in a repeated phrase of T4. This indicates that the perceptual dominance of the TTLS melody in voice perception was weakened. Each variation included cue phrases such as “Up above the world…,” predicting the recurrence of the TTLS melody (Supplementary Figure 2). The two musical structures—variation and ternary forms—establish global and regional contexts, respectively. In each variation, T4 is introduced after the second iteration of the cue phrase within the regional ternary context, facilitating anticipation of melodic recurrence. Moreover, at the global level of the variation form, the same structure involving T4 is repeated in Variation II and Variation IV, further enhancing anticipatory processing. The training of the repeated upper voice and its perceptual prominence might have facilitated participants' recognition of the lower voice (Taher et al., 2016). The LTDMI value was higher in Variation IV than in Variation II during T4. We interpret that the connectivity reduction in Variation II for T4 is attributable to the properties of its lower voice, which differed from them in Variation IV (Figure 3). Our findings show that participants did not solely focus on the TTLS melody at T4 but could also detect other sounds.

In our previous study (Kim et al., 2020), unidirectional transmission of information from the left IFG to the right HG, showing frontotemporal connectivity, was associated with the recognition of the TTLS melody in the heading of each variation. However, the involvement of temporofrontal connectivity, which was in the opposite direction to the frontotemporal connectivity, was observed in repeated phrases in musical context evoked (Figure 2B). The temporofrontal connectivity was strongly correlated with the frontotemporal connectivity in T4 (Figure 2B and Supplementary Table 2). The top–down processing of a familiar melody can significantly influence the figure–ground relationship (Strüber and Stadler, 1999; Nelson and Palmer, 2007). Considering that the roles of frontotemporal and temporofrontal connectivity are linked to the TTLS melody-based and sensory-based processes, respectively, this heightened connectivity could indicate a dual process: extracting the novel lower voice in the target phrase, including the familiar TTLS melody, and dissecting the components within the novel lower voices. The temporofrontal connectivity supports the frontotemporal connectivity. Thus, bidirectional connectivity of the frontotemporal and temporofrontal pathways between the left IFG and the right HG is possibly modulated by both a top–down process based on knowledge of the TTLS melody and a bottom–up process based on new information on the voices accumulated while sequentially listening to target and cue phrases in each variation (Alho et al., 2015; Dzafic et al., 2021).

Numerous researches on the auditory figure–ground relationship have conducted auditory scene analysis (Bregman, 1994) and grouping (Wagemans et al., 2012) using tasks that discriminate a sound pattern as a figure from a ground of tone and chord sequences with irregularities in spectral and temporal properties (Teki et al., 2011; O'Sullivan et al., 2015; Toth et al., 2016). Functional magnetic resonance imaging and electroencephalography studies (Teki et al., 2011; O'Sullivan et al., 2015; Toth et al., 2016) have reported that regions involved in discriminating this figure–ground perception comprise the primary auditory area, superior temporal sulcus, superior temporal gyrus, intraparietal sulcus, medial/superior frontal gyrus, and cingulate cortex. The processing of multiple voices in this study involved the IFG and the HG, both exhibiting information transmission. The left IFG is crucial in processing familiarity (Plailly et al., 2007) and in music-syntactic processing, indicating implicit learning (Sammler et al., 2011) and contributing to memory retrieval (Platel et al., 2003; Watanabe et al., 2008). Moreover, the left IFG is involved in differentiating between melody and accompaniment (Spada et al., 2014). The left IFG is highly activated during the recognition of pattern deviations in melody (Habermeyer et al., 2009), musical novelty in a context (Tillmann et al., 2003), and conscious experience (Weilnhammer et al., 2021). The left IFG is also associated with sematic and syntactic processing (Zhu et al., 2022), and sentence comprehension (van der Burght et al., 2019). In contrast, the right HG predominantly processes tone deviance (Sabri et al., 2006; Nan and Friederici, 2013) and the segregation of auditory streams (Snyder et al., 2006). The right auditory cortex is the dominant site for music processing (Perani et al., 2010), auditory stream segregation (Snyder et al., 2006), and spectral pitch (Schneider et al., 2005).

The IFG and HG are pivotal areas for music perception, and their connectivity is discussed in relation to syntax processes (Papoutsi et al., 2011; Kim et al., 2019, 2021), categorization (Roswandowitz et al., 2021), and the working memory of the melody process (Burunat et al., 2014). The temporofrontal network is engaged in the categorization of vocal signals (Roswandowitz et al., 2021), while frontotemporal connection is involved in both top-down and bottom-up processes in sensory learning (Dzafic et al., 2021). Directional information flows within frontotemporal connectivity may therefore explain how the left IFG and the right HG collaborate in processing target phrases. The involvement of the left IFG and right HG may reflect the entire process of naturally grasping, comparing, and understanding the voices in target phrases within a particular context rather than simply perceiving them as sounds. Accordingly, enhanced frontotemporal and temporofrontal connectivity may reflect the integrated processes by which the brain recognizes the TTLS melody based on memory, segregates the melody, and detects differences in the lower voices relative to the prior context.

Previous studies have selectively manipulated stimuli or directed participants' attention to specific auditory streams (Uhlig et al., 2013; Ragert et al., 2014; Spada et al., 2014; Strait et al., 2015; Hausfeld et al., 2018; Puschmann et al., 2019; Barrett et al., 2021). Attention has been shown to be critical for figure–ground perception (Poort et al., 2012). Therefore, research on such perception uses artificially composed stimuli to direct participants' attention. During the MEG experiment in this study, all participants passively listened to the naturalistic music of Mozart's 12 Variations, K. 265, without any instructions regarding focusing their attention on a specific voice or melody. Thus, whether both figure and ground were processed attentively or pre-attentively remains unclear, given the absence of intentional attention control. Although listeners may focus on a particular voice while listening to music, the changing flow of music can encompass brief perceptible moments in which the figure–ground relationship continuously shifts without listeners consciously realizing it. Indeed, music listeners can automatically process information such as syntactic errors and tone deviations without intentional attention (Maess et al., 2001; Naatanen et al., 2007). We interpreted that participants could attentively or pre-attentively detect sonic changes in the voices at that moment, although our data do not confirm that non-musicians could perceptually segregate the streams or identify which voice evoked the sonic differences (Figure 3). Our results successfully captured the moment when the figure–ground relationship between the upper and lower voices changed, as evidenced by the difference in frontotemporal connectivity for repeated phrases in the two variations and the correlation between frontotemporal and temporofrontal connectivity.

Familiarity is critical for explaining the figure–ground experiment (Palmer, 1999; Hulleman and Humphreys, 2004; Nelson and Palmer, 2007). The target phrases in our stimuli involved the TTLS song, which has been used in studies related to the perception of familiar melodies (Trehub et al., 1985; Upitis, 1990; Besson et al., 1994; Creel, 2019). Familiarity would be naturally implied in the theme and all of its variations, considering that Mozart's 12 Variations, K. 265, is based on the TTLS melody. Logically, the familiarity implied in the TTLS melody might influence participants' figure–ground perception. However, in our previous study focusing on the TTLS melody (Kim et al., 2020), we could not directly prove the effect of familiarity on frontotemporal connectivity as the connectivity changed irrespective of the presence or absence of the TTLS melody. In our present study, the same TTLS melody appeared repeatedly in Variations II and IV. The effects of familiarity are consistent in both Variations II and IV. Thus, the TTLS melody was used to assess changes in the figure–ground relationship.

Naturalistic stimuli have been used to examine various topics (Saarimäki, 2021; Izen et al., 2023; Tervaniemi, 2023). In studies on the concepts of emotion (Singer et al., 2016; Putkinen et al., 2021), melodic expectation (Kern et al., 2022), temporal aspects of rhythm and beat (Sturm et al., 2015; Weineck et al., 2022), and familiarity (Leaver et al., 2009), multiple naturalistic pieces have been used as musical stimuli. Some studies using naturalistic stimuli have examined their hypotheses on topics such as motif, musical features, timbre, and depression, based on a single piece (Alluri et al., 2012; Cong et al., 2013; Burunat et al., 2016; Liu et al., 2020). Our hypothesis was also created for the melody of TTLS and the figure–ground relationship of voices using Mozart's 12 Variations, K. 265. In the fields of audiation (Uhlig et al., 2013; Ragert et al., 2014; Hausfeld et al., 2018; Barrett et al., 2021) and vision (Peterson et al., 1991; Lamme, 1995; Super et al., 2003; Zhang and Von Der Heydt, 2010; Von der Heydt, 2015), changes in the figure–ground relationship between the elements of an object can be simply observed and explained by comparing the related objects. However, naturalistic music has its own narrative, which can be described by its structure. The connectivity reflects complicated processes for the target phrases of 2.1 s without the context and for each 2.1-s-long target phrases in the theme and variations, leading up to the target phrase. This approach, however, may constitute a critical shortcoming of our study, compared with measurements using artificially composed stimuli. Furthermore, no verbal reports or other measures were obtained during or after MEG recording to determine what participants perceived as the “figure” in the music at each moment, including critical time windows. As a result, connectivity changes were measured under naturalistic listening conditions in which participants passively listened to music. These limitations can be addressed in future studies using novel paradigms that incorporate detailed behavioral responses and larger sample sizes. Nevertheless, our results reflected the humans' ubiquitous experiences with individual participants.

In addition to the use of naturalistic music, our study had other limitations. While non-musicians can perceive the figure separately from the ground (Toth et al., 2016), they tend to focus more on the upper voice than on a lower one (Sloboda and Edworthy, 1981). In contrast, musicians are more sensitive to voice perception and are better able to distinguish voices (Fujioka et al., 2005; Strait et al., 2015). Music training significantly influences selective attention (Puschmann et al., 2019). We did not recruit musicians as participants to examine our hypothesis in terms of basic musical ability. Recruiting non-musicians might have impacted our results. The temporofrontal connectivity may show a statistically significant distinction (P < 0.05) since the recognition of different elements in the lower voices (via temporofrontal connectivity) involves a more complex cognitive process. Therefore, future studies should verify these results with musicians. Furthermore, this study concentrated on frontal and temporal regions. Our findings should be verified at the whole-brain level. In the current experimental paradigm, we did not consider an additional test individual preference of subjects. The preference is also influential of music listening, and which should be addressed importantly in further studies with a novel experimental paradigm. The MEG recording and analysis approach used in this study may be replicable using other modalities, such as EEG. Despite these limitations, the use of Mozart's 12 Variations, K. 265, was invaluable in understanding the fundamental neural processes associated with processing real music featuring multiple voices and elucidating the human experience of music. Our findings elucidated how the brain dissects voices from the multidimensional structures of music and reconstructs the figure–ground relationship between voices.