Imagine looking for someone in a crowd, trying to keep the person's characteristics in mind while suppressing other, potentially distracting events: Constantly bombarded with a continuous stream of sensory information, our brain needs to select, filter and prioritize: the use of top-down processes for this task is indispensable. Recent research suggests that neural oscillations, rhythmic fluctuations in the excitability of neural populations, are the brain's key feature in these processes: Events that coincide with the oscillation's high excitability phase are amplified whereas events occurring during the low excitability phase are suppressed, and the brain seems to use this mechanism as a powerful tool to gate and filter input (Schroeder and Lakatos, 2009). This mechanism can also be seen as a way of environmental subsampling: “Snapshots” of the environment are taken at a rate that corresponds to the frequency of the respective oscillation and the moment of the “snapshot” might be optimized by an alignment of neural oscillations with external events (for a review, see VanRullen et al., 2014). Moreover, the oscillatory power can impact the overall responsiveness of a given brain region, a mechanism that has been associated with a modulation of the neural firing rate (Haegens et al., 2011b; Jensen et al., 2012).

An important role of neural oscillations for attentional selection and stimulus processing¹ has been shown across modalities: For the visual (Lakatos et al., 2008), auditory (Stefanics et al., 2010), somatosensory (Haegens et al., 2011b), motor (Arnal, 2012), and olfactory systems (Kay, 2014). Although the basic mechanisms, common across modalities, are relatively well understood (Schroeder and Lakatos, 2009; Arnal and Giraud, 2012; Calderone et al., 2014), there seem to be differences in oscillatory mechanisms of stimulus selection between modalities whose systematic investigation began only recently (Thorne and Debener, 2014; VanRullen et al., 2014). In this paper, we will contrast the two modalities that are arguably the most important for human perception and behavior: vision and audition. Recently, it has been suggested that these modalities are confronted with different requirements for stimulus processing, largely due to fundamental differences in the input the two systems receive: Whereas visual input is relatively stable in time and might not require processing that is precise on a millisecond time scale, auditory input changes rapidly and relies crucially on a processing system that can cope with fast-fluctuating information (Thorne and Debener, 2014; VanRullen et al., 2014; Zoefel et al., 2015). Here, we go one step further and summarize and discuss differences in the oscillatory mechanisms underlying stimulus processing and selection in vision and audition. We argue that these differences are a direct consequence of the requirements imposed on each system by the particular input. We start by giving an overview of oscillatory frequencies involved in stimulus processing and selection in the two systems (Section “Frequencies of Stimulus Processing: Summary”). In the core of this article (Section “Relation to the System's Input”), we then structure these findings systematically, based on different properties (timing, predictability, and salience) of the stimulus input, and on consequences of these properties for oscillatory processes. This section is guided by several questions: Can the two systems adapt to their environment—and do they even need to? Do oscillatory mechanisms depend on whether the stimulus is rhythmic (arguably the preferred case for oscillatory processing as an alignment between oscillation and stimulus is possible) or only a single event (Section “Adjustment vs. Entrainment”)? What happens when the input is unpredictable or unattended (Section “Processing “modes””)? Answering these questions has critical implications for our understanding of neural oscillations involved in attention and stimulus selection. As we will see, significant progress has been made in recent years, but new questions arise with the increased knowledge. Those questions are also addressed in this paper. Several hypothetical answers are provided, based partly on previous findings and partly, as we emphasize here, on speculation. Experimental approaches that are necessary to investigate the proposed hypotheses are also discussed.

There is overwhelming evidence for the alpha band (7–13 Hz) as the principal frequency range of stimulus processing in the visual system (Figure 1). This observation was already published by Berger (1929) who reported a dependence of alpha power on the visual input: Alpha power in the electroencephalogram (EEG) increases when subjects close their eyes. Since then, both theoretical and experimental approaches provided convincing evidence that the alpha band is related to an inhibition (or disengagement) of brain regions (Klimesch et al., 2007; Jensen and Mazaheri, 2010; Foxe and Snyder, 2011): For instance, alpha power increases in the hemisphere that is ipsilateral to an attended stimulus (and therefore less strongly involved in its processing) (Thut et al., 2006; Sauseng et al., 2009), or in brain regions not involved in the current task (Zumer et al., 2014). Moreover, it has been shown that visual perception is directly related to the alpha band: The detection of a visual target depends on alpha power (Hanslmayr et al., 2007; Romei et al., 2008; Figure 1A). EEG alpha phase impacts both the probability of detecting a visual target and the likelihood of perceiving a phosphene during transcranial magnetic stimulation (TMS) (Busch et al., 2009; Mathewson et al., 2009; Dugué et al., 2011; Figure 1B), and random visual input seems to reverberate in the brain at a frequency corresponding to the alpha band (VanRullen and Macdonald, 2012; Figure 1C). Similarly, when systematically testing a wide range of physiologically plausible frequencies, the strongest neural resonance in response to rhythmic visual input (e.g., as steady-state response) is observed in the alpha band (Herrmann, 2001; de Graaf et al., 2013), and a longer-lasting manipulation of neural activity by electric current has mostly been reported in that frequency range (e.g., an increased power can be observed several minutes after the stimulation; Thut and Miniussi, 2009; Zaehle et al., 2010). A single pulse of TMS induces a reverberation of endogenous alpha oscillations, but of no other frequency bands (Herring et al., 2015). Together, these findings might indicate that the intrinsic frequency of neurons and/or neuronal circuits (Hutcheon and Yarom, 2000) in the visual system is indeed located predominantly in the alpha band. Finally, both the probability of detecting a visual stimulus after a cue (Figure 1D) and the following reaction time fluctuate periodically (Landau and Fries, 2012; Fiebelkorn et al., 2013; Song et al., 2014). In these studies, the perceptual and behavioral fluctuations have been found at a frequency of 4 Hz per visual hemifield, with the two 4 Hz rhythms in opposite phase, indicating an overall rhythmicity of 8 Hz, and thus lying within the alpha band (Zoefel and Sokoliuk, 2014).

Following recent work by VanRullen (2016), one important distinction should be made here: Whereas some studies report effects in the alpha band around 10 Hz, linked to a topographical distribution that is centered on the occipital lobe (e.g., Figures 1A,C), the peak frequency of the effect described in other studies seems to be somewhat lower and located in more frontal² regions (7–8 Hz; e.g., Figures 1B,D). Indeed, a systematic compilation of different studies investigating the role of EEG phase for perception yielded prominent effects at two different frequencies, 7 Hz and 11 Hz (see Figure I in VanRullen, 2016). In a recent study, Keitel and Gross (2016) applied sophisticated signal analysis methods to resting-state magnetoencephalography (MEG) data in order to characterize the spectral profile (termed “spectral fingerprints”) measured in different brain regions. Interestingly, they demonstrated a clear 10 Hz (but no 7 Hz) peak in occipital regions, and a 7 Hz (but no 10 Hz) peak in Inferior Frontal Gyrus. It is thus likely that the two types of effects stem from different generators of oscillatory processing (VanRullen, 2016), a point that we will return to in the following sections. Also, it is unclear whether a frequency of 7–8 Hz can be assumed to reflect “textbook alpha” (or whether it is rather part of the theta band)—nevertheless, for the sake of simplicity, in the following, we will designate both bands as “alpha,” but differentiate between an “occipital alpha” (~10 Hz) and “frontal alpha” (~7–8 Hz).

Although neural activity in the gamma band (~30–70 Hz) has often been reported in the visual system, gamma band power might be tightly linked (“coupled”) to the phase of the alpha band (Bahramisharif et al., 2013; Roux et al., 2013; Jensen et al., 2014): Indeed, the 8-Hz periodicities observed in visual detection performance seem to be correlated with changes in gamma power that fluctuate at the same rhythm (Landau et al., 2015). Gamma activity is often associated with bottom-up processing of sensory information and is present across sensory systems (Fontolan et al., 2014; Bastos et al., 2015). In this paper, we focus on slower frequency bands associated with top-down components of stimulus processes (e.g., attentional selection or predictions) and refer instead to comprehensive reviews published on gamma oscillations in the brain (Fries et al., 2007; Ray and Maunsell, 2015).

The dominant frequency of stimulus processing in the auditory system is less clear than in the visual one (Figure 2): On the one hand, many studies describe an alignment between the phase of neural oscillations in the delta/theta band (~1–8 Hz) and rhythmic stimulation (Lakatos et al., 2008; Schroeder and Lakatos, 2009; Stefanics et al., 2010) and this alignment can decrease reaction time (Stefanics et al., 2010), increase efficiency of stimulus processing (Cravo et al., 2013) and seems to be present even after stimulus offset (Lakatos et al., 2013; Hickok et al., 2015). Phase entrainment can also be observed when subjects do not consciously perceive the stimulus, ruling out contamination by evoked potentials (Zoefel and Heil, 2013; Figure 2B).

On the other hand, the alpha band seems to be important as well (Obleser et al., 2012; Strauß et al., 2014; Weisz and Obleser, 2014): Alpha power can be modulated by auditory attention like in the visual system (Kerlin et al., 2010; Frey et al., 2015), speech intelligibility co-varies with alpha power (Obleser and Weisz, 2012; Wöstmann et al., 2015), and the phase of the alpha band modulates auditory stimulus detection if entrained by transcranial alternating current stimulation (tACS; Neuling et al., 2012). Moreover, the power of the gamma band can be coupled to most frequency bands (Lakatos et al., 2005; Fontolan et al., 2014; however we note that, to our knowledge, alpha-gamma coupling has yet to been shown in the auditory system).

Although the auditory system seems to “resonate” (e.g., as steady-state response to a rhythmic stimulus) most strongly in the 40-Hz (i.e., gamma) range (Galambos et al., 1981), several studies suggest that similar phenomena can be found in lower frequency bands as well (e.g., Liégeois-Chauvel et al., 2004). Moreover, human auditory perception is most sensitive to amplitude fluctuations and frequency modulations at a frequency of ~4 Hz. This has been demonstrated in a multitude of psychophysical experiments using a wide range of stimuli (e.g., amplitude- or frequency-modulated tones or noise) and measures (e.g., discrimination thresholds), and has been summarized extensively by Edwards and Chang (2013). Thus, it is difficult to determine a distinct frequency of stimulus processing in the auditory system (indeed, there are “spectral fingerprints” at many different frequencies in Superior Temporal Gyrus, the location of auditory cortices; Keitel and Gross, 2016). Instead, the auditory system might utilize different frequencies for different purposes, and the reported results have to be interpreted in tandem with the respective stimulation protocol, as argued in the following section.

Table 1 summarizes the proposed contrasts between the visual and the auditory system in terms of oscillatory mechanisms involved in stimulus processing and selection. Some properties might be common across all systems: Neural oscillations can be used as a tool for attentional selection, and both oscillatory power and phase can be used to gate stimulus input. Changes in power might reflect a tonic suppression of processing (e.g., in a region that is currently not involved in stimulus processing) and/or change the effectiveness of the phase of an oscillation, cycling between moments of amplification and suppression. In the absence of attention, an “alpha-mode” (“occipital alpha” in the visual system) seems to be present in both systems, and is associated with a state that is decoupled from external stimulation and in which only very salient events can overcome the increased alpha amplitude and reach consciousness. Indeed, a recent study by Haegens et al. (2015) demonstrated alpha band activity in primary cortical regions of the macaque across all modalities.

However, there are differences between the visual and auditory systems: The oscillatory entrainment to rhythmic stimulation seems to be a fundamental feature of the auditory system, probably evolved due the rhythmic nature of the auditory environment. Indeed, the tendency to synchronize with auditory rhythms is ubiquitous: We sing, we dance, we clap in response to music or even to a simple beat (Nozaradan, 2014). Importantly, this phenomenon is much less pronounced for the visual system: For instance, the urge to dance is significantly lowered when watching someone dancing without the corresponding sound. Thus, although in principle the visual system also seems to be able to entrain, the adjustment of power and phase might be more important in this system—visual stimuli are often predictable, but rarely rhythmic. Interestingly, and in line with this notion, it has been shown that the auditory system is superior to the visual one when movement has to be synchronized with a rhythmic sequence in either or both modalities (Repp and Penel, 2002; Patel et al., 2005; Merchant et al., 2015) and auditory rhythmicity can influence the perceived flicker rate of a visual stimulus but not vice versa (Shipley, 1964; Herrmann et al., 2015). Task-irrelevant information in the auditory system impairs visual processing more strongly than vice versa if this information is of temporal nature (Guttman et al., 2005). Thus, although visual stimuli can in principle influence auditory processing and perception (potentially using alpha oscillations; Thorne et al., 2011; van Wassenhove and Grzeczkowski, 2015) and do so even more prominently if rhythm is involved (Ten Oever et al., 2014), a multitude of findings indicates that the auditory system dominates the visual one in the time domain (an extensive summary of the literature on this conclusion is provided in Grahn, 2012). Finally, a simple cue (without rhythmic component involved) is sufficient to introduce the mentioned periodic fluctuations in visual performance (Landau and Fries, 2012; Song et al., 2014; Zoefel and Sokoliuk, 2014), and recent findings suggest that the latter are present even in the absence of stimulation (Busch et al., 2009; Busch and VanRullen, 2010; Landau et al., 2015). The situation seems to be different for the auditory system, where similar periodic fluctuations in performance have been reported only after the offset of a rhythmic stimulus (Hickok et al., 2015) and possibly cannot be observed in the absence of rhythmicity (e.g., after a single cue; Zoefel and Heil, 2013; VanRullen et al., 2014).

In the presence of attention, stimulus processing in the visual system might be focused on the (“frontal”) alpha band (see Box 1), irrespective of the frequency of stimulation, whereas the auditory system adapts its dominant frequency of processing to that of the environment. Speculatively, if stimulation is non-rhythmic, the auditory system might operate in the alpha rhythm as well (see Box 2). However, whereas the visual alpha rhythm(s) might subsample sensory regions even when stimulus timing is unpredictable, it is possible that the auditory alpha is decoupled from sensory processes; in this way, the auditory system can avoid a loss of information that occurs when subsampling is applied to rapidly fluctuating auditory information with unknown timing. An alternative during unpredictable event timing in the auditory system would be a “continuous mode” of stimulus processing in which most oscillatory sampling mechanisms are suppressed. We note that “alpha mode” and “continuous mode” might even co-exist, but at different stages of auditory processes, or depend on the level of vigilance (see Box 2).

Neural oscillations are a powerful tool of the brain to prioritize and select relevant information while ignoring distracting input. This article summarizes the current state-of-the-art and provides several proposals that can be systematically tested and extended. Future studies and theories are indispensable to advance this exciting field of research.

All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.