Temporal aspects of music—the organization of sounds into patterns and the unfolding of those patterns over time—allow us to predict what and when sounds are likely to occur next. It is supposed that such expectations and predictions are crucial for the experience of pleasure during music listening (Salimpoor et al., 2015). Thus, the question of how we track the temporal structure of music is key for understanding why we love music.

On a basic level, musical rhythms can be defined as acoustic sequences with patterns of duration and accentuation; they represent the structure of music (London, 2004; Large et al., 2015). We can tap our feet or bob our heads in time with music because we are usually able to perceive a regular beat or pulse within a rhythm. If we perceive strong and weak beats, detect regularly recurring events, and use these events to group a fixed number of beats in cycles, we can be said to have found the meter of a musical piece (London, 2004). Broadly speaking, a quadruple meter encourages us to tap our feet four times (beat-level) in each cycle (bar-level), whereas a triple meter encourages us to tap three times in each cycle.

Polyrhythms consist of at least two different beat-levels (N and M) within one bar-level, whereby N and M have to be relatively prime (i.e., they have no common roots; Large and Kolen, 1994). Polyrhythms, such as 3:2, 4:3, 5:4, or 7:5, create rhythmic tension and are used in different genres including, but not limited to jazz (Pressing, 2002), rock, metal (Pieslak, 2007; Osborn, 2010), African drumming (Locke, 1982), and twentieth century Western art music (Poudrier and Repp, 2013). Described from a perspective of embodiment, polyrhythms can be seen as contrasting but coordinated body movements (Iyer, 2002). For example, let us imagine a 3:2 polyrhythm. A dancer listening to this polyrhythm might stamp his or her feet three times each cycle, aligned with the triple beat, while in the same amount of time moving his or her upper body once to the left and once to the right in time with the duple beat.

Various bimanual tapping studies predominantly conclude that contrasting pulse trains in simple polyrhythms are not processed by parallel timekeepers, but integrated into one shared rhythmic framework (e.g., Klapp, 1979; Deutsch, 1983; Klapp et al., 1985, 1998; Jagacinski et al., 1988; Summers and Kennedy, 1992; Summers et al., 1993; Peper et al., 1995; Krampe et al., 2000; Kurtz and Lee, 2003). However, at least in professional musicians, bimanual tapping of faster polyrhythms might be controlled by independent timekeepers (Krampe et al., 2000). Even without overt movements, integrated processing of contrasting patterns can be beneficial when tracking polyrhythms (e.g., Jones et al., 1995; Klein and Jones, 1996). Jones et al. (1995) examined attentional strategies of musicians and non-musicians when trying to detect timing changes of target tones in 3:2 polyrhythms. In a series of experiments that encouraged either integrative or selective attending, the authors found that although musicians as compared to non-musicians were more accurate in detecting timing changes, they were not better at adjusting their attentional strategy (i.e., choosing integrated vs. selective attending).

Whereas, most of the aforementioned studies used simple polyrhythms consisting of two contrasting isochronous pulse trains, Poudrier and Repp (2013) examined whether classically trained musicians use divided or selective attention when keeping track of polyrhythms of different complexities in a perceptual probing paradigm. In simple polyrhythms the two contrasting parts (2/4 and 6/8 meter) could be tracked simultaneously. However, more complex polyrhythms limited the musicians' ability to divide their attention. Keller and Burnham (2005) found similar results in a dual-task paradigm that examined the recognition and reproduction of matched-metrical and mismatched-metrical rhythmic patterns. In a recent study, Wöllner and Halpern (2016) asked expert and novice conductors and pianists to detect pitch and timing deviations of target tones in selective- and divided-attention tasks (focus on one of two melodic streams, and focus on both streams, respectively). They found that in both tasks experts detected more targets than students. Furthermore, conductors—specialists in monitoring multiple parts of a musical piece simultaneously—detected more targets than pianists in the divided-attention task.

Training-related differences were also shown in sensorimotor synchronization studies. Finger tapping studies show that musical training can sharpen rhythm perception and production (e.g., Chen et al., 2008; Cameron and Grahn, 2014; Matthews et al., 2016). Another tapping study by Drake et al. (2000b) suggests that musicians as compared to non-musicians have a better apprehension of metric structure and can process a wider range of hierarchical levels. These behavioral results are supported by EEG studies examining event-related potentials (ERPs) elicited by target tones within musical rhythms. In isochronous sequences of unaccented tones that typically induce binary subjective accenting (“tick”–“tock”–“tick”–“tock”…), ERPs differed more strongly between deviants that fell on subjectively strong and subjectively weak beats for musicians as compared to non-musicians (Brochard et al., 2003). Another study found that, in musicians but not in non-musicians, mismatch negativities (MMN: an ERP difference measure that is increased by unexpected stimuli) were elicited by violations of numerical regularity in tone sequences (van Zuijen et al., 2005). Additionally, the difference between MMNs elicited by meter-congruent and meter-incongruent target tones was larger in musicians as compared to non-musicians (Geiser et al., 2010).

The enhanced ability of musicians in the processing of hierarchical temporal structures might relate to an increased involvement of the working memory, as reflected by an enhanced recruitment of the dorsolateral prefrontal cortex (Chen et al., 2008). Musicians' advantages in rhythm perception and performance can be linked to the tight coupling between sensory and motor areas of the brain that develops through musical training and experience (Haueisen and Knösche, 2001; Bangert and Altenmüller, 2003; Bangert et al., 2006; D'Ausilio et al., 2006; Grahn and Rowe, 2009; Stupacher et al., 2013). Although, previous research has shown differences between musicians and non-musicians in sensorimotor synchronization, perceptual organization of rhythm, and neural processing, it remains an open question whether rhythm-related neural oscillations are modulated by musical expertise and in which ways (Nozaradan, 2014).

Neural oscillations are based on synchronized fluctuations between high and low excitability states in groups of neurons (Singer, 1999; Fries, 2005), a function that is, among others, important for attention and prediction (Buzsáki and Draguhn, 2004; Schroeder and Lakatos, 2009; Arnal and Giraud, 2012; Frey et al., 2015). In music, meter describes the percept of a regular temporal structure with alternating strong and weak events. Resonance theories link pulse and meter perception with neural oscillations (Large, 2008). They assume that when neural oscillations entrain to a stimulus with at least some temporal regularity, such as most music, high neural excitability states and the underlying pulse of the musical rhythm become aligned (Large and Kolen, 1994; Large and Jones, 1999; Large, 2008; Schroeder and Lakatos, 2009). When measured with EEG, such entrained oscillations are called steady-state evoked potentials (SSEP) and are reflected in amplitude peaks in the frequency spectra of ongoing signals. In a series of studies Nozaradan and colleagues showed that beat-related SSEPs are influenced by the imagination of accents within unaccented isochronous pulse trains (duple vs. triple meter imagery; Nozaradan et al., 2011), are present with rhythmic patterns whose envelope spectra do not necessarily have maximum peak amplitudes at the beat frequency (Nozaradan et al., 2012a), and relate to sensorimotor coupling (Nozaradan et al., 2015). Whereas, Nozaradan and colleagues used abstract and relatively simple auditory stimuli, Tierney and Kraus (2015) showed that neural entrainment also plays a key role in tracking the rhythm of real music.

In a previous experiment we used pop/rock drum rhythms and isochronous pulse trains to show that, during listening, stimuli with more complex rhythmic hierarchies, as compared to isochronous stimuli, elicited more pronounced SSEPs at the beat's second harmonic frequency (Stupacher et al., 2016). In the same experiment, auditory stimuli were interrupted by silent breaks (2–6 s), allowing us to examine whether neural oscillations persist without external stimulation as well as to connect ongoing neural oscillations (SSEPs) and transient neural responses (ERPs). Similar to the SSEP results, the N1 ERP component indicated that listening to more complex rhythms increased neural entrainment. Although, we were not able to directly show neural oscillations in the frequency domain during silent breaks, N1 results suggested that beat-related oscillations persist throughout the breaks, but become blurred and/or decrease in their amplitude with increasing time. This is an important finding, as it suggests that beat-related neural oscillations are not merely a reflection of the auditory stimulus (bottom-up processing), but rather involve top-down endogenous timing processes.

The silent break design indirectly instructed the participants to imagine that the rhythm continued during silence. Gordon (1985, 1999) coined the term audiation, which describes the hearing of music in the mind. Audiation involves the interpretation and understanding of what was just heard, recalled, composed or performed. Several neuropsychological studies indicate that the auditory cortex can be active without the actual presence of sound and suggest that musical imagery is related to this activity (Zatorre and Halpern, 2005). Audiation, or more generally musical imagery, is an important part of music education (e.g., Gordon, 1999). A recent study showed that musical training and higher musical engagement were associated with better performance in a musical tempo imagery task (Jakubowski et al., 2016).

In the current experiment, we decided to adapt the silent break design of Stupacher et al. (2016). Musicians and non-musicians listened to auditory stimuli that started with an isorhythmic quadruple drum rhythm and turned into a 3-over-4 (4:3) polyrhythm. After the polyrhythm, we included a silent period of ~2 to 3 s after which a target stimulus was presented. Participants had to decide whether this target stimulus fell on the timing grid of the triple-beat part of the 4:3 polyrhythm or was presented 220 ms too early or 220 ms too late. This design allowed us to address various open questions: How do neural oscillations evolve during listening to stimuli with changing metrical hierarchies? Do neural oscillations persist without an acoustic input when the task demands to imagine that the beat goes on? And does musical expertise modulate neural entrainment?

Empirical findings, computational models, hypotheses and theories suggest that the entrainment of neural oscillations to musical rhythm enables us to disentangle and organize temporal structures and to perceive a steady beat (Large et al., 2015). The current study adds to this growing body of research. By using auditory stimuli that changed from quadruple isorhythms to 3-over-4 polyrhythms, we showed that neural oscillations evolve with changing and overlaying rhythmic frameworks. By adapting a previous design in which the stimuli were interrupted by silent breaks (Stupacher et al., 2016), we provided evidence that endogenous neural oscillations persist without external stimulation. By combining EEG recordings and behavioral data, we assessed the relevance of endogenous neural entrainment for rhythm processing. And finally, by comparing data of musicians and non-musicians we gained first insights into whether beat-related neural oscillations are modulated by musical expertise and in which ways.

The neural resonance theory of pulse and meter proposes that neural oscillations entrain to the temporal patterns of music (Large and Kolen, 1994; Large and Jones, 1999; Large, 2008). Neural entrainment requires a neural oscillator, a periodic external stimulus, and the synchronization between the two (Thut et al., 2011; Large et al., 2015). All of these assumptions were met in the current study. The time-frequency plots in Figures 8, 9 show the evolvement of neural entrainment over time during listening to the different parts of the auditory stimuli. As proposed by the neural resonance theory of pulse and meter (Large and Kolen, 1994; Large and Jones, 1999), the plots show that neural oscillations entrained to the different metric hierarchies of the rhythms. During the isorhythmic quadruple beat (0–4 s), neural oscillations of both, musicians and non-musicians are strongest at 2 and 4 Hz (quarter-note and eighth-note level of quadruple beat, respectively). During the subsequent 4:3 polyrhythm (4–8 s) the patterns change: Neural oscillations at 2 and 4 Hz become weaker and additional oscillations at 1.5 and 3 Hz emerge. This observation of fine-resolved oscillatory dynamics is supported by more commonly used analyses of the EEG signals' frequency spectra (Table 2, Figures 5, 6), which showed that quadruple rhythms elicited SSEPs at 2 and 4 Hz and that 4:3 polyrhythms additionally elicited neural oscillations at 1.5 and 3 Hz.

Figure 6 shows that the quadruple rhythms (Figure 6A) and the triple-beat metronome (Figure 6D)—the two different parts of the 4:3 polyrhythm—resulted in contrasting SSEP patterns. Each pattern reflects the metric hierarchies of the corresponding auditory stimulus. SSEPs of the 4:3 polyrhythm (Figure 6B) appear to be a combination of quadruple- and triple-beat-related SSEPs. This is an interesting finding, because participants were instructed to focus on the triple beat of the polyrhythm. Additionally, the behavioral task made it advantageous for the participants to only track the triple beat. If the two different parts of the polyrhythm could be processed independently by parallel timekeepers, participants should have focused their attention on the triple-beat part. Consequently, with parallel timekeepers the frequency pattern of neural oscillations during the polyrhythm should reflect the triple-beat part and be similar to the frequency patterns during the silent period (Figure 6C) and the triple-beat metronome (Figure 6D). However, in both musicians and non-musicians, the most pronounced SSEP can be found at the quadruple-beat-related frequency of 2 Hz. Additionally, the frequency spectra of both musicians and non-musicians show a prominent peak at 6 Hz—the only frequency shared by the quadruple and triple beat (Figures 5B, 6B). Importantly, the frequency spectra of the sound envelopes of the polyrhythms (Figure 2B, bottom panel) do not show such a prominent peak at 6 Hz. In line with previous findings, these results suggest that polyrhythms, at least up to a certain complexity, are not processed by parallel timekeepers but rather within a shared rhythmic framework (e.g., Deutsch, 1983; Klapp et al., 1985, 1998; Jagacinski et al., 1988; Summers and Kennedy, 1992; Summers et al., 1993; Jones et al., 1995; Peper et al., 1995; Klein and Jones, 1996; Krampe et al., 2000; Kurtz and Lee, 2003). Further evidence for a shared framework comes from the time frequency plots that revealed a prominent ERP elicited by the second onset of the triple-beat woodblock and a less prominent ERP elicited by the fifth triple-beat onset (marked by dotted lines in Figures 8, 9). The second and fifth onsets of the triple beat woodblock are the first onsets that lie outside the metric grid of the preceding quadruple beat (in the first and second bar of the 4:3 polyrhythm, respectively). Thus, these onsets are less expected than the others, because they violate the previously established metric structure. At first glance this finding seems to contradict the assumption of a shared rhythmic framework. However, as both the amplitudes in the time-frequency plots and the amplitudes of the N1 ERP components indicate that the second onset of the triple beat woodblock elicited more prominent responses than the fifth onset, it can be concluded that the fifth onset was more expected than the second onset. This finding suggests that the triple beat is gradually integrated into the rhythmic framework of the quadruple beat. Consistent with this interpretation, the third and sixth onset of the triple beat woodblock did not elicit an observable ERP.

In general, the finding of evolving neural oscillations that entrain to rhythmic stimuli with changing and overlaying rhythmic frameworks shows that the analysis of SSEPs provides a useful tool for the investigation of polyrhythm processing. Future EEG or EMG experiments could use the frequency-tagging approach to investigate whether more complex polyrhythms are processed differently (e.g., in parallel streams).

Besides neural entrainment to physically present acoustic rhythms, we recorded and analyzed endogenous neural entrainment during the silent periods between the presentations of rhythmic stimuli and target stimuli (Figure 1B). In musicians, the pattern of endogenous neural oscillations (Figure 6C, left panel) resembles the pattern observed during listening to the triple-beat metronome (Figure 6D, left panel). In non-musicians (Figure 6C, right panel), oscillations at 1.5 Hz stand out. The normalized peak amplitude at 1.5 Hz for both groups combined was larger than zero [t₍₂₅₎ = 3.02, p = 0.006], indicating that endogenous neural oscillations at this frequency persisted throughout silent periods. As expected, these findings suggest that participants successfully imagined that the triple beat continues during silent periods. In line with this conclusion, Figure 7 shows that during the silent periods, task-relevant endogenous neural oscillations at triple-beat-related frequencies (1.5 and 3 Hz) were more pronounced than task-irrelevant oscillations at quadruple-beat-related frequencies (2 and 4 Hz). This is an important finding, as it suggests that neural oscillations during silent periods are not representing a “washing out” effect of the previous stimulation. If the endogenous neural oscillations were based on such a “washing out” effect, amplitudes at triple-beat- and quadruple-beat-related frequencies should be equally high.

The finding of beat-related neural oscillations that persist without external stimulation indicate that neural entrainment to music is not only driven by bottom-up processes, but also by top-down processes and endogenous oscillatory networks. Thus, our results support conclusions of a previous study using the silent break design (Stupacher et al., 2016), and studies showing that beat-related neural oscillations are elicited even if the fundamental beat frequency is not present in the acoustic signal (Nozaradan et al., 2012a, 2016; Large et al., 2015). In contrast to our previous study (Stupacher et al., 2016), which did not show persisting beat-related neural oscillations after participants listened to a drum rhythm, the current study provides direct evidence for endogenously created neural oscillations in a listening task without overt beat-related movement. A plausible reason for this discrepancy is that the instructions lead to different mental imagery states. Part of the task of the current study was to try to imagine that the triple beat continues during the silent period, whereas in Stupacher et al. (2016), participants were not instructed or explicitly motivated to imagine that the beat goes on after the stimulus stopped. By explicitly instructing participants to imagine that the stimulus continues, the current experiment intensified the recruitment of top-down processes, probably leading to strengthened endogenous neural entrainment.

Additional evidence for endogenous neural entrainment comes from the analysis of ERPs elicited by target tones that were presented after the silent periods (Figures 1B,C). For correctly identified target stimuli, the P1 amplitude elicited by targets presented on the triple beat timing grid (on time) was lower than the P1 amplitude elicited by too early and too late targets. We interpret the lower P1 amplitude as reflecting higher prediction accuracy because correctly identified on time targets evoked significantly lower P1 components than incorrectly identified on time targets. Support for this interpretation comes from a study by Moberget et al. (2008) who showed an enhancement of P1 amplitudes elicited by aperiodic stimuli (tones with mixed 650, 800, and 950 ms inter-stimulus-intervals) compared to periodic stimuli (fixed 800 ms inter-stimulus-interval) in healthy control participants. The functional relevance of the ERP data for timing judgements is supported by the finding that P1 amplitudes and P2 latencies were only affected by target time when the subsequent answer in the behavioral task was correct. How musical imagery relates to the ERP findings, why the amplitudes of N1 and P2 components were not modulated by the target time, how the effect of target time on P2 latencies can be explained, and what the underlying mechanisms of the P1 modulation could be remains a matter of future work.

By including a behavioral task, we were able to assess the functional relevance of endogenous neural entrainment (i.e., beat-related neural oscillations in silent periods) for rhythm and beat perception. Participants correctly identified the timing of the target stimulus (too early, on time, or too late) in about 51% of the cases (33.3% chance level), indicating that the task was not too easy and not too hard. Musicians tended to give more correct answers than non-musicians. In line with this finding, Rammsayer and Altenmüller (2006) showed that musical training can increase the sensitivity to deviations from regular beats. Interestingly, the difference between musicians and non-musicians was mainly driven by responses to early target stimuli (Figure 4). Overall, correct identification was highest for early and lowest for late targets. Late targets were often perceived as on time targets (Figure 4B). This results was unexpected, as in comparable conditions of a study by Barnes and Jones (2000; e.g., Table 2, 600 ms standard duration) correct identifications of shorter, same, or longer intervals compared to a context sequence were more symmetrically distributed. When comparing the duration of two time intervals, the preceding context rhythm can affect the temporal judgements (e.g., Barnes and Jones, 2000; McAuley and Jones, 2003). How the context of polyrhythms affects temporal judgements of a subsequent target tone remains to be investigated in greater detail.

For the whole sample of participants, we could show that endogenous neural entrainment improves fine-grained temporal processing, as indicated by a positive correlation between the number of correctly identified target stimuli and the amplitude of neural oscillations at 1.5 and 3 Hz during silent periods. This finding suggests that listeners use dynamic temporal patterns in form of entrained neural oscillations to process musical rhythms and to predict rhythmic events. In a recent study, Nozaradan et al. (2016), came to similar conclusions by showing that SSEPs during listening to an unsyncopated rhythm positively correlate with synchronization accuracy while tapping in time with the beat of the same rhythm. In the same study the authors computed a measure of endogenous neural entrainment by subtracting the SSEP amplitude during listening to the unsyncopated rhythm from the SSEP amplitude during listening to a syncopated rhythm. Comparable to the directly measured endogenous neural entrainment in the current study, the more indirect measure computed by Nozaradan et al. (2016) was associated with better temporal predictions.

To date, most neuroimaging studies on neural entrainment to music neglected the fact that musical expertise shapes how we perceive the temporal structures of music. In the current study, the comparison between musically trained and untrained participants revealed two main findings.

First, neural entrainment did not differ between musicians and non-musicians during listening to musical rhythms with low (triple metronome), medium (isometric quadruple rhythm), and high (4:3 polyrhythm) rhythmic complexities. These null results are intriguing because various behavioral and neuroimaging studies have shown that musical training sharpens temporal processing (e.g., Drake and Botte, 1993; Rammsayer and Altenmüller, 2006; van Vugt and Tillmann, 2014), improves sensorimotor synchronization accuracy and stability (e.g., Franek et al., 1991; Drake et al., 2000a,b; Krause et al., 2010; Repp, 2010; Baer et al., 2013; Cameron and Grahn, 2014; Baer et al., 2015), deepens the comprehension of metrical structures (e.g., Drake et al., 2000a,b; Brochard et al., 2003; van Zuijen et al., 2005; Geiser et al., 2010), modulates patterns of neural activation during rhythm perception, and strengthens neural auditory-motor coupling (e.g., Haueisen and Knösche, 2001; Bangert and Altenmüller, 2003; Drost et al., 2005; Haslinger et al., 2005; Vuust et al., 2005; D'Ausilio et al., 2006; Bangert et al., 2006; Lahav et al., 2007; Grahn and Rowe, 2009; Stupacher et al., 2013). The question arises as to why neural entrainment during stimulus presentation did not differ between musicians and non-musicians, although all of the aforementioned advantages of musical training seem to be relevant for entrainment processes.

To answer this question we should first take a look at the second main finding from the comparison of musical trained and untrained participants: Endogenous neural entrainment, as measured in silent periods, was stronger in musicians compared to non-musicians. This effect was shown by multiple different analyses. First, the amplitude of endogenous neural oscillations at triple-beat-related frequencies was significantly larger than zero in musicians but not in non-musicians. Second, in musicians, the amplitude of endogenous neural oscillations at triple-beat-related frequencies was significantly larger than the amplitude of task-irrelevant oscillations related to the quadruple beat (Figure 7). In non-musicians, this effect only showed a slight tendency in the same direction. Third, in musicians, the pattern of endogenous neural oscillations during silent periods (left panel of Figure 6C) resembled the pattern of neural oscillations elicited during listening to the triple metronome (left panel of Figure 6D). In contrast, the pattern of non-musicians (right panel of Figure 6C) lacked endogenous neural oscillations at frequencies other than 1.5 Hz. Consistent with this observation, the difference of endogenous neural entrainment between musicians and non-musicians was mainly driven by the strength of neural oscillations at 3 Hz—the second harmonic of the task-relevant triple beat. Fourth, the time-frequency plots of Figure 8 showed that endogenous neural oscillations (8–10 s) at triple-beat-related frequencies (1.5 and 3 Hz) were more pronounced in musicians compared to non-musicians. And finally, in accordance with the results of the neural entrainment analyses, behavioral results showed that musicians tended to be better at correctly identifying the timing of the target stimulus presented after the silent period.

To sum up the effects of musical expertise, analyses of ongoing neural activity and behavioral results suggest that endogenous neural entrainment during silent periods was stronger in musicians compared to non-musicians. In contrast, no difference of neural entrainment between the two expertise groups was found during stimulus presentation. How can we explain this divergence? The endogenous neural entrainment measured during silent periods is mainly driven by top-down processes, as participants had to imagine the triple beat without any external stimulation. In contrast, the neural entrainment measured during listening to the rhythms reflects a combination of top-down and bottom-up processes (Nozaradan et al., 2016; Stupacher et al., 2016), whereby the bottom-up component is driven by the acoustic input. During listening to music as well as during music production, hard-wired bottom-up sensation and experience-based top-down perception are interdependent. For example, one and the same rhythmic pattern can be interpreted in different ways depending on the metric framework the listener uses to structure the pattern (Honing, 2012). The different interpretations can be a result of different listening experiences, musical training backgrounds, or other individual factors. When briefly exposed to music of a foreign culture (2 weeks), infants, in contrast to adults, developed a culture-specific responsiveness and were able to differentiate culture-specific rhythmic patterns (Hannon and Trehub, 2005). Put another way, long time exposure to culture-specific music can narrow the types of musical patterns that we can effortlessly process. The stronger endogenous neural entrainment in musicians might also relate to better musical imagery abilities of musicians or musically engaged individuals compared to less musically trained or engaged individuals (e.g., Aleman et al., 2000; Jakubowski et al., 2016). Although top-down processes are crucial for beat and meter perception, our findings suggest that the amplitude of endogenous neural oscillations is lower than the amplitude of bottom-up driven neural oscillations. Thus, musical expertise might strengthen top-down driven endogenous neural entrainment during rhythm perception, but hard-wired bottom-up processes that do not differ between musicians and non-musicians might cover this effect.

In conclusion, the current study provides new evidence showing that neural oscillations at beat-related frequencies persist after the external acoustic input stopped. This finding indicates that besides bottom-up stimulus-driven processes, neural entrainment to music involves endogenously created neural oscillations related to higher order cognitive processes that enable us to predict what and when sounds are likely to occur. The comparison of musicians and non-musicians suggests that those endogenously created neural oscillations are strengthened by musical expertise. Moreover, our study demonstrates that neural entrainment is crucial for rhythm perception and processing, even when listening to stimuli with complex rhythmic frameworks such as polyrhythms.

JS and MW designed the study, managed data collection, analyzed the data, and wrote the manuscript. GW designed the study and wrote the manuscript.