Appropriate studies were identified on the basis of predetermined inclusion and exclusion criteria, designed using the PICOS framework, as listed in Table 1.

We performed the literature search using three databases: PubMed, Web of Science and Scopus. All terms were searched for within the title, abstract and keywords field and were structured as follows:

(“Zwicker tone”

OR

((“auditory OR “hearing”)

AND

(“afterimage” OR “after image” OR “after-image”)))

Hand searching was also carried out following screening. Reference lists of included records were combed to identify any studies that may have been omitted by our initial search strategy. The abstracts of identified studies were read, and if deemed potentially suitable were added straight to the full-text screening stage.

After removal of duplicates, all retrieved records were screened independently by two reviewers (JLRB and JAS), using the predetermined inclusion and exclusion criteria. Screening was performed using Covidence software. Any discrepancies during independent screening were resolved via discussion and consensus between the two reviewers. Initial screening was performed based on title and abstract, after which full texts were retrieved for remaining references and then rescreened. Exclusion of papers at the full text screening stage required justification based on eligibility criteria, provided by both reviewers independently. Discrepancy either in the full-text inclusion verdict, or exclusion reasoning were again resolved through discussion and consensus.

The table used for data extraction was designed collaboratively via independent trialling, comparison and consensus by two reviewers. Upon satisfaction with the table and approach, data extraction was performed by one reviewer. During data extraction, snowballing was performed: reference lists of included papers were hand-searched for any other potentially relevant citations not captured by the initial search strategy. Any that appeared suitable upon reading their abstract were added straight to the full-text screening stage to be assessed for inclusion. Where graphical data was extracted, the ‘Plot Digitizer’ tool (https://plotdigitizer.com/app) was used to aid precision. Where notch widths were converted to equivalent rectangular bandwidths (ERBs) the following equation was used: ERBs = 21.4 log10 (0.00437·f + 1), where f represents frequency (Moore and Glasberg, 1983).

Following our search strategy we retrieved 162 records. After removal of duplicates, 80 records underwent title/abstract screening, from which 22 papers were identified for full-text screening. Seven papers were excluded having not fulfilled the eligibility criteria, including three for which full texts could not be retrieved. 15 records remained of which one was re-reporting data from another included study and was subsequently also excluded. Hand searching during data extraction identified two more eligible studies, giving a total of 16 records (Figure 1). It should be noted that one included record (DeGuzman, 2012) was a master’s thesis rather than a peer-reviewed paper.

All records bar one (Zwicker, 1964) reported the age of their subjects to some extent; with mean ages consistently falling between 20 and 30 years old, though some ranges include subjects up to 50 years old (see Table 2). The majority of studies (86%) made some allusion to the hearing thresholds of their participants, with nine reporting audiograms having been performed. Where hearing condition was reported, all subjects were of normal hearing according to the threshold determined by the authors, though said definition ranged from <10 dB 250–8000 Hz to self-reported assessment.

Stimuli employed were most commonly notched noise, though in some cases low-pass noise was used to induce the percept (see Figure 2; Table 3). Fastl et al. (2001), in addition to low-pass and notched noise, included noise signals with pure tones embedded, either within broadband noise, or at the edge frequency of spectrally contrasted noise which reportedly also generate an illusory tone.

In the case of notched stimuli, the centre frequency used ranged from 500 Hz to 8.1 kHz, though these extreme values were primarily used in studies exploring ideal parameters, with most studies primarily using centre frequencies between 1–6 kHz (see Figure 3). Several studies (Chen et al., 2025; Fastl and Stoll, 1979) have failed to induce a ZT at low notch centre frequencies (250, 500 and 1,000 Hz). While Chen et al. (2025) did not induce a ZT with a 1 kHz centre frequency notch (5.2 ERBs, 3 s), Fastl and Stoll (1979) did (4.1 ERBs, 60s). This likely reflected the increase in noise duration (from 3 to 60s) rather than the modest difference in ERB width. Generally, it seems that the lowest centre frequency at which the Zwicker tone can be induced likely lies between 1 and 2 kHz for short duration noise, though this could be reduced by varying other parameters such as notch width or noise duration.

Reported notch widths range from 0.8–17.4 ERBs, though again these extreme values again appear to be outliers, both only appearing in one paper each, with the majority of widths tending to within 3-7ERBs. Leske et al. (2014) attempted to find the most effective parameters, and their ranges have been re-employed by studies which have subsequently reported the notch frequency/width combination most effective at inducing ZT perception: Qi et al. (2022) report this as being a 1 octave (6.1ERB) notch centred at 4 kHz, while Mohan et al. (2020) it as a 0.77 octave (4.7ERB) notch centred at 3.7 kHz.

In addition to notch width and centre frequency other parameters also appear to modulate the probability of perception and perception strength. Lummis and Guttman (1972) varied spectral-gap depth (the relative spectrum in the notch versus the upper and lower noise bands) and the noise duration. They found that for short duration noise (less than 2–3 s) a large gap-depth (i.e., slope of the notch in the frequency domain) is required becoming increasingly large as duration reduces. Above 2–3 s the gap depth becomes less of a factor and stimuli of most durations/gap depths (above 30 dB) produce a reliable percept. The same authors also suggest that spectrum level of the noise is a factor where 10-30 dB spectrum level represents a sweet spot for inducing the ZT, whereas at higher spectrum levels the percept is less likely to be perceived.

There is considerable variation in the reported prevalence (see Table 4) of listeners who do perceive Zwicker tones (from hereon we shall refer to this population as ZT⁺), as opposed to those who do not (ZT⁻). Zwicker (1964) originally stated that 90% of naive subjects could perceive his auditory after-image, with this statistic increasing to 100% upon prompting. Lummis and Guttman (1972) report a similar figure, 96% following notched noise, along with 37% when only high-pass noise is used. However, more recent studies suggest a significantly lower value: ranging between 30 and 52% (Parra and Pearlmutter, 2007; Ueberfuhr et al., 2017; Chen et al., 2025).

While only five studies present ZT prevalence as a binary statistic, i.e., grouping subjects into ZT⁺ and ZT⁻, three further studies instead grouped participants into more than two categories representing the likelihood of perception over a scale. While we cannot confidently convert these categories into an equivalent binary ZT⁺ and ZT⁻ statistic, we can extract the reported ZT⁻ figure (classified in these studies as those individuals least likely to respond positively to ZT stimuli) and from it infer an upper-bound to the ZT⁺ statistic. These values range from 11 to 32%, suggesting a maximum ZT prevalence of between 68 and 89%. Admittedly, this statistic cannot be used to confirm the more recent, lower values of ZT prevalence, but combined with said values they do cast doubt on the very high ZT prevalence originally reported by Zwicker (1964) and Lummis and Guttman (1972). Finally, ZT perception has been repeatedly linked to tinnitus perception but, to date, only one study has empirically tested this potential link. Parra and Pearlmutter (2007) found a prevalence of 91% in individuals with tinnitus compared to 42% in individuals with normal hearing.

In the reported assessments and surveys establishing ZT prevalence, the inclusion of a white noise control stimulus is common, with some records also employing a non-ZT inducing notched noise as a further control. Where ZT prevalence is being calculated, control stimuli are mostly used to pre-screen participants based on a threshold set by the authors, before the remaining participants are organised based on their responses to notched noise (Qi et al., 2022; Chen et al., 2025; Mohan et al., 2020). The thresholds for exclusion in these studies vary between 50–90% correct rejection of the white noise. However, DeGuzman (2012), determined perception based on accuracy calculated using both hits (responding positively to ZT inducing stimuli) and correct rejection.

Zwicker’s original observation that the frequency of the percept falls within the range of the notch of the inducing stimulus has been consistently reproduced in situations where notched noise is employed. In support of this, five studies report data from frequency matching tasks in which a quiet tone is adjusted to match the perceived pitch of the Zwicker tone. Gockel and Carlyon (2016) having investigated ZT-pitch matching in musically trained individuals, finding that, as a general rule, the frequency of the percept is roughly 1.1–1.2 times the frequency of the lower edge of the notch (i.e., the edge frequency of the low frequency noise band). Interestingly, another study reported ZT⁺ participants perceive the Zwicker tone following low-pass noise and the addition of an upper noise band has little effect on frequency (Lummis and Guttman, 1972). This suggests that the pitch of the Zwicker tone is strongly influenced by the edge frequency of the lower band of noise. However, while the higher noise band does not seem to alter the ZT pitch an increase in notch width, and hence a higher upper edge frequency, seems to result in a wider range of notch widths that produce a ZT (Lummis and Guttman, 1972) and greater perceived ZT loudness (Leske et al., 2014; Lummis and Guttman, 1972).

Where studies performed frequency matching, we extracted mean values, converted them into ratios of perceived ZT frequency to lower edge frequency (LEF) of the inducing notched noise. This reveals a range of 1.12–1.34 times the lower edge frequency of the inducing notched noise (Table 5). This is a wider range than predicted by Gockel and Carlyon (2016), however some variation in the ratio should be expected not least because, as previously stated, the intensity of the inducing stimuli is a known factor: higher SPL stimuli induce higher pitched Zwicker tones. Interestingly, both Zwicker (1964) and Wiegrebe et al. (1996) explore the effect of increasing SPL and report a similar increase in ZT pitch (ZT/LEF ratio) as a function of SPL: +0.06/10 dB and +0.10/10 dB increase, respectively. To explore this relationship between percept frequency and sound level, we plotted our extracted ZT pitch as a function of SPL, revealing a linear positive correlation across the data points from these studies (r (7) = 0.83, p = 0.006, Figure 4). Furthermore, when considering the lower edge frequency as the informant of the percept frequency, Wiegrebe et al. (1996) varied both the notch width and intensity and found a much greater effect of intensity than notch width on the frequency multiplication factor.

Three papers report changes in perception of tones presented following notched noise in ZT⁺ participants. Presentation of a notched, ZT-inducing noise precursor can reduce detection threshold for a pure tone presented just after noise offset, both versus a broadband (ie. without a notch) noise precursor (Wiegrebe et al., 1996; Norena et al., 2000), and versus detection in silence (Wiegrebe et al., 1996). The reported magnitude of this effect varies: a maximum mean improvement of 8.1 dB is reported for vs. silence (Wiegrebe et al., 1996), while improvements vs. broadband noise range from 2.2–7.2 dB (Wiegrebe et al., 1996; Norena et al., 2000) – all at 40 dB SPL noise presentation. In addition to notched noise, low-pass noise (also ZT-inducing) sees a similar though lesser effect versus broad band noise (Norena et al., 2000). Interestingly, detection thresholds are reduced at frequencies within the notch, but for those frequencies proximal to the spectral edges of the noise an increase in thresholds is reported (Norena et al., 2000). Again, a similar effect is reported for low-pass noise, with an increase in detection threshold at the noise edge and a reduction in threshold for tones above the edge frequency. This benefit to detection threshold is then lost at higher frequencies, returning to baseline as the frequency becomes more distant from the noise edge (Norena et al., 2000).

As with ZT perception, detection threshold improvements vs. silence are modulated by the width of the notch and the level of the noise, with wider notches and quieter noise presentation reportedly resulting in lesser threshold reductions (Wiegrebe et al., 1996). However, only one comparison has been made of each: 30–40 dB and 3–6.1ERBs, and as such this relationship may vary over greater ranges, likely peaking at some value for either parameter before reducing again rather than improving indefinitely. One study also reported ZT frequency matching and found that maximum threshold improvement to be similar to the reported frequency of the ZT (Wiegrebe et al., 1996). Furthermore, Norena et al. (2000) report threshold improvement exclusively in the ear ipsilateral to ZT perception suggesting the two phenomena may be related. However, all participants in said study reportedly experience a monaural percept, in contrast to previous findings that suggest binaural fusion of the percept after binaural presentation (Lummis and Guttman, 1972).

Finally, in addition to changes in absolute thresholds, perceived loudness of a subsequently presented pure tone is also reportedly coincidental with ZT perception at low presentation levels (5–10 dB above absolute threshold), with ipsilaterally presented tones perceived as louder following notched vs. broadband noise (Norena et al., 2002). Again, this effect is seen at a frequency within the notch, while the opposite effect is seen outside of the notch with tones perceived as quieter. At higher presentation levels (>10 dB above threshold), no difference is reported in perceived loudness either within or outside of the notch.

We identified six records that report data from cortical electrophysiology studies (five EEG and one MEG), and various neural correlates of Zwicker tones are presented. The heterogeneity of methods, stimuli and results make comparison complicated, though below we attempt to synthesise these studies. That said, the experimental designs used can be, broadly, categorised into two groups. (1) within-subject designs: trials in which a participant responded positively to a stimulus are compared to those trials in which they do not report ZT perception. (2) between-subject designs: trials are compared between ZT⁺ and ZT⁻ groups for ZT-inducing and non-inducing stimuli.

An increase in ERP amplitude is reported in ZT⁺ subjects vs. ZT⁻ following ZT inducing stimuli (between-subject design), around 500-900 ms post noise offset (Chen et al., 2025; Mohan et al., 2020). This result has been seen both where the ZT-inducing stimulus was homogenous (Chen et al., 2025) and bespoke (Mohan et al., 2020) for each subject. Conversely, when using a within-subject design, DeGuzman (2012) found much earlier ERP differences, one around 140 ms and one around 340 ms when ZT⁺ participants responded positively to ZT-inducing stimuli vs. negatively. These differences were not observed in ZT⁻ subjects, i.e., when comparing their ZT positive trials (responded yes to a ZT) versus negative trials. Chen et al. (2025) also observed a significant correlation between evoked ERP amplitude and both the probability of perceiving a ZT and subjective ZT intensity.

Oscillatory power within the theta band (3-8 Hz, as defined by the included studies) has been shown to be increased generally (i.e., total theta power) and following ZT-inducing stimuli (i.e., induced theta power) in ZT⁺ vs. ZT⁻ subjects (Qi et al., 2022; Chen et al., 2025; Mohan et al., 2020). Of these, two studies found an increase in induced theta between 200 and 900 ms (Qi et al., 2022; Mohan et al., 2020). Whereas, Chen et al. (2025) found an increase in induced theta power between 168 and 296 ms but not during the later 600 and 900 ms epoch. Average total theta power (averaged across significant channels during the significant epoch) scales significantly with ZT percept strength (Chen et al., 2025; Mohan et al., 2020) and the probability of perceiving a ZT (Chen et al., 2025). In addition, single-trial theta power also significantly predicts the ZT percept strength (Mohan et al., 2020).

Alternatively, Leske et al. (2014) found reduced oscillatory power in the alpha/beta range (10-20 Hz) when contrasting neural responses to ZT inducing vs. non-ZT inducing stimuli (within-subjects design) in ZT⁺ subjects. As with the above studies this neural correlate was found to scale significantly with the strength (loudness) of the ZT (Leske et al., 2014). Meanwhile, DeGuzman (2012) did not report a significant effect of post-stimulus offset (ZT related) alpha and did not find a significant change in ongoing alpha (during the notch noise) in ZT⁺ subjects. However, they did observe an increase in frontal alpha on trials in which ZT⁻ subjects responded positively, reporting ZT-perception, during the ongoing stimulus.

Source localization of ERPs does not appear to be consistent across this limited number of studies (three records). Mohan et al. (2020) suggest sources in the temporo-parietal junction, right parietal and auditory cortex. Whereas, the same group subsequently found increased activation in right medial orbitofrontal cortex, right rostral middle frontal gyrus and left lateral occipital sulcus (Chen et al., 2025). In addition, DeGuzman (2012) observed an early negativity (~140 ms) on fronto-central channels and the later positivity (~340 ms) on central-parietal channels. Neural correlates of the ZT have been reported as represented tonotopically, with equivalent current dipole depth increasing with ZT frequency (Hoke et al., 1998), though this has not been replicated. Source localisation of theta power has, simialrly to ERP source localization, been inconsistent across studies. Interestingly, the ZT percept can be amplitude modulated by transcutaneous electrical stimulation, similar to real tones (Ueberfuhr et al., 2017).

Our results show great variation in the reported prevalence of ZT perception, ranging from 30–100% when presented as a binary statistic. As stated in our results, in cases where perception is categorised into more than two groups, using the ZT⁻ statistic to infer an upper limit of ZT prevalence suggests that Zwicker (1964) and Lummis and Guttman (1972) estimates of 96–100% may be too high and more conservative estimates may be more accurate. A range of possible factors might explain the differences in observed ZT prevalence.

For example, the stimuli used may have an influence on the prevalence observed. Zwicker (1964) used 60s noise bursts to induce the ZT, far longer than the standard 2–3 s duration utilised in the literature. Lummis and Guttman (1972) employed a more typical signal duration of 4 s, however this was cycled with 1 s off durations and the majority of subjects reported ZT perception within 2 min, though some only weakly. In contrast, Ueberfuhr et al. (2017) report a perception probability of only 52% having used a 2 s on-off cycle: they do not report maximum lengths of time allowed for participants to potentially perceive the ZT, but they do state that the stimulus sequence could be restarted at any time. It may be that the ZT is elusive for many people and initially unperceivable, but extended presentation allows the percept to appear, albeit faintly, however more recent uses of extended, cycled stimuli suggest this may not be the sole cause for increased perception probability.

The participants themselves may present a factor in this disparity: Zwicker (1964) omits any demographic information about his subjects, though it is not unreasonable to suspect that they may well have been lab colleagues. Meanwhile, Lummis and Guttman (1972) state that their initial survey was completed by ‘randomly selected laboratory personnel, most of whom were naive listeners’. Though said listeners were reportedly naïve, the distinction between lay people and auditory laboratory personnel may well be important, the latter presenting a group of potential expert listeners.

Finally, the methodology of these earlier papers is potentially less robust than more recent studies that include control conditions and false alarms: rather they simply ask whether a percept is experienced or ask participants to describe their perception. The lack of controls make it difficult to assess whether participants were indeed detecting the tone reliably, though the inclusion of frequency matching provides some confidence in the findings. Furthermore, the question used to probe whether listeners heard the percept - i.e. asking listeners whether they heard a soft tonal sound between noise bursts - while maintaining naivety regarding the source of the percept (indeed, participants are reportedly surprised to learn that the ZT is not externally delivered), does prompt the expectation of a tone, perhaps doubly confounding in the absence of a control condition. Although, it should be noted almost all studies will have informed participants that they are listening for a soft tone after noise cessation, i.e., prompting listeners expectation. There is some support for the idea that prompting/priming might increase ZT prevalence. Zwicker (1964) reported that prevalence increased upon prompting and description of the percept: “For only two listeners was it necessary to discuss in detail the sound that they should listen for. They may have expected too much, and after the discussion they were able to detect and to match the pitch of the subjective sound in the same way as the other 18 people.”

A range of parameters appear to be important for ZT induction. Overall, the general features of the Zwicker tone presented are consistent with Zwicker’s original report – a tonal, illusory percept, experienced after the offset of broad band noise containing a spectral notch or low frequency broadband noise (though a notch is more likely to cause the percept and makes it louder). The centre frequency of the notch and the notch width both also modulate the probability of perception as does the duration of inducing stimuli. When presented monaurally, it is only perceived in the ear ipsilateral to presentation; binaural, diotic presentation leads to central, fused perception; and dichotic presentation leads to two distinct, lateralised percepts. The percept increases in duration with a longer induction stimulus, and a 100 ms inducer presented at a repetition frequency of 5 Hz can produce an ongoing percept. The frequency of the percept falls within the region of the notch, and an increase in inducer sound level leads to an increase in percept frequency, though increasing the sound level too high may reduce or prevent the percept (Lummis and Guttman, 1972). In addition, when calculating percept frequency as a ratio of lower edge frequency and plotting it as a function of sound level, our results support the suggestion that the frequency of the ZT is related to the lower edge frequency of the inducing stimuli. This also suggests that sound level is broadening auditory filers to push the percept higher in frequency.

The approach to categorising an individual as ZT⁺ or ZT⁻ might also benefit from examination. As noted, the use of control stimuli in the form of white noise or non-ZT inducing notched noise, is common and arguably essential. Indeed, both Zwicker (1964) and Lummis and Guttman (1972) omit a control stimulus in their initial surveys, which may contribute to their higher reported ZT prevalence.

Currently both binary, ZT⁺ vs. ZT⁻, and more continuous categorisation, ZT perception on a scale, are used. The potential difficulty associated with ZT detection, as suggested by the higher prevalence associated with longer noise exposure, may make an argument for a more continuous categorisation, i.e., rating subjects on the likelihood of their perception rather than assigning them a binary label. If indeed the salience of the percept varies between individuals on a continuous scale, it stands to reason that said salience would be similarly continuous as a function of stimulus parameters. Multiple studies have explored the use of a range of parameters in order to idealise stimuli or tailor them to individual participants (Qi et al., 2022; Leske et al., 2014; Mohan et al., 2020), however the distribution of percept rating values has not been reported in detail. Supplementary data from Qi et al. (2022), demonstrates that when given a 0–9 scale, participants tend to respond in a primarily binary fashion, demonstrating a preference for extreme ratings of 0 or 9, suggesting that perception may be closer to a binary effect. Indeed, those studies that employ a range of parameters and report prevalence still all identify groups who reliably do not perceive the ZT (Qi et al., 2022; Mohan et al., 2020).

The idea that ZT perception is, largely, binary (a person can be either ZT⁺ or ZT⁻) is also somewhat supported by individual data. In order to highlight this we make use of signal detection theory (Macmillan NA, 2002). Signal detection theory is a framework for separating out the sensitivity of a listener to a given signal (e.g., a real or Zwicker tone) from their bias to respond a given way (Macmillan NA, 2002). For example, we often present a signal (e.g., a pure tone) at a given intensity (e.g., sound level) a set number of times and calculate the proportion of times the participant correctly detects the signal, this is known as the “hit rate.” This measure, however, presents a problem: the participant might respond “yes I heard the signal” on every trial, i.e., they are biased toward reporting that they detected a signal. These individuals cannot be considered particularly good at the task but based on hit rate alone they appear to be very sensitive to the stimuli. To account for this, we include stimuli both with and without a signal (positive and negative stimuli, respectively). The proportion of trials a participant indicates the presence of a signal in response to a negative stimulus is called the “false alarm rate.” Some participants might consistently respond “yes” (regardless of the stimulus), they will have both a high hit and false-alarm rate. Conversely, some participants might have a moderate hit rate but a very low false alarm rate: these people detect the signal less frequently but, crucially, rarely identify a negative stimulus as containing a signal. Based on hit rate we could, wrongly, conclude the former participants were very good, when they are just biased toward respond “yes,” at the task and the latter not very good, when in fact they are they are just biased toward responding “no.”

Signal detection theory offers a method for reducing the influence of a person’s bias to measure a person’s sensitivity to a given stimulus using the d prime (d’) statistic. We calculate d’ by taking the difference between hit rate and false alarm rate (as z scores): a low to moderate d’ (e.g., 1–2 or above) is considered relatively sensitive and a value below this relatively insensitive. As well as sensitivity, we can use signal detection theory to measure decision criterion, which is independent of the actual sensitivity to the stimuli, rather reflecting an individual’s bias to a certain response (Macmillan NA, 2002).

Most studies use a non-ZT inducing sound (wideband noise or non-ZT inducing notch noise) as a control stimulus to check for false alarms and so we can, where data is available, apply signal detection theory to these data. Two of these studies (DeGuzman, 2012; Chen et al., 2025) reported individual hit and false alarm scores, allowing us to retroactively calculate d’, and plot individual data points on a receiver operating characteristic plot (Figure 5). The vast majority of participants fell into two clusters (Cluster 1 (top-left), close to: hits = 1, false alarms = 0; Cluster 2 (bottom-left), close to: hits = 0 false alarms = 0) suggesting that on the whole most participants fall into one of the two groups, Cluster 1 comprising ZT⁺ individuals and Cluster 2 comprising ZT⁻ [while Chen et al. (2025) used a selection criterion that might skew there data into two groups, there is no evidence DeGuzman (2012) does the same]. These data also suggest that, on the whole, the classification criteria used (while different for these two studies), did a reasonable job of separating individuals into ZT⁺ and ZT⁻ groups. However, Figure 5 also demonstrates the difficulty in using criteria (based on hits and false alarms) to exclude/include participants. Participants grouped as “threshold” by DeGuzman (2012) (grey diamonds) showed similar sensitivity (e.g., d’ > 2) to participants categorised as ZT⁺. Conversely, Chen et al. (2025) classify a participant with moderate sensitivity (d’ ~ = 1, small filled circle on d’ = 1 line) as ZT⁺ but two others with moderate sensitivity as ZT⁻ (small empty circles around 0.4 hits and above d’ = 1 line). We propose that it may be more suitable to use either a d’ based threshold to group people into ZT⁺ and ZT⁻ groups or a statistical threshold based on single trial data to reduce the occasional inconsistencies in categorisation that might occur.

The question of whether Zwicker tone perception is binary or on a continuum remains open. A range of stimulus parameters can modulate whether a Zwicker tone will be perceived. For example, the best notch for inducing ZT perception is close to 4 kHz with an ~1 octave notch (see “Induction Stimuli” section). Increasing the notch width from 0.6 to 1 octave (notch centred on 4 kHz) increases the probability of detecting a ZT (Qi et al., 2022). Lowering the centre frequency of the notch (but keeping it at 1 octave) reduces the probability of detecting a ZT (Chen et al., 2025). In addition, the duration of the inducing stimuli may also alter the probability of detecting a ZT (see “Prevalence” section). Finally, given altering sound level modulates the frequency of the percept and the likelihood a tone will be detected in the frequency range of the ZT it also seems likely it will modulate the likelihood a ZT is perceived. While the current, though hardly extensive, data suggests ZT perception is binary rather than on a continuum studies employing a range of stimulus parameters (mapping out a large range of parameters) would be required in order to determine whether a person labelled as ZT⁻ is truly unable to hear ZTs under any circumstance or merely hears them over a restricted range of parameters.

ZT induction stimuli are associated with improved detection thresholds of subsequently presented tones in silence, in comparison to both broadband noise (Norena et al., 2000) and silence (Wiegrebe et al., 1996). Surprisingly, Wiegrebe et al. (1996) reported improvements are greater vs. silence than broadband noise, though this may be due to the different methodological approaches used to establish either threshold: Békésy tracking system for silence (Wiegrebe et al., 1996) and a 2-alternative forced choice paradigm for broadband noise (Wiegrebe et al., 1996; Norena et al., 2000).

Changes in detection threshold following notched noise are also reported outside of the ZT literature. The finding that notched noise presentation can improve signal detection of threshold vs. silence has been reported in animals exposed to long term notched noise, with improvements again seen within the frequency range of the notch and not outside (Krauss and Tziridis, 2021). Interestingly the authors also report that 91% of animals exposed to notched noise also developed tinnitus like behaviours at the frequency of the notch. Of those animals demonstrating tinnitus behaviour, 40% recovered post noise exposure when their hearing essentially experienced full recovery. Improvements vs. broadband noise are also confirmed in the literature (Alves- Pinto and Lopez- Poveda, 2008; Larsby and Arlinger, 1998), however the finding that narrower notches result in greater threshold reduction vs. silence is reportedly not the case vs. broadband noise, where wider notches see greater improvements (Larsby and Arlinger, 1998). The report that threshold improvement is greater at louder noise presentation (Wiegrebe et al., 1996) — at least for 30 dB vs. 40 dB - has been explored elsewhere and found to peak at 70 dB, reducing either side towards 50 dB and 90 dB (Alves- Pinto and Lopez- Poveda, 2008).

The finding that the frequency of the ZT is closely related to the frequency of greatest detection threshold improvement (Wiegrebe et al., 1996) has not been further confirmed within the literature due to the lack of ZT-frequency matching in other studies that measure threshold change across a range of frequencies (Norena et al., 2000; Alves- Pinto and Lopez- Poveda, 2008). On the basis of our plotting the ratio of percept to lower edge frequency as a function of presentation level (Figure 4), we can estimate the frequency of a ZT that might hypothetically arise following the notched noise presentation and compare this with reported frequencies of greatest threshold change. Norena et al. (2000) presented a notch with a lower edge frequency of 3.4 kHz at 40 dB SPL and subsequently we would predict the frequency of greatest threshold change to be 3.95 kHz. Indeed, they find the greatest improvement at 4 kHz, the tested frequency closest to the prediction. Meanwhile, Alves- Pinto and Lopez- Poveda (2008) utilise a notch with lower edge frequency 7 kHz ranging between 50 and 90 dB SPL presentation level. At 50 dB we would predict the frequency of greatest change to be 8.46 kHz and Alves-Pinto report 8.5 kHz. This comparison of reported values versus prediction can be seen in Figure 4. At higher presentation levels (70 dB+) our predictions are less accurate, consistently returning higher frequencies than are reported. However, it should be noted that a relatively narrow notch is used, between 7 and 9 kHz, and at 70 dB + we would predict the frequency of greatest change to be within the upper band of noise, which may explain the reduced reliability, at least in part.

A number of mechanisms have been proposed to account for the generation of the ZT percept including habituation (Fastl et al., 2001), gain adaptation (Parra and Pearlmutter, 2007) and stochastic resonance (Schilling et al., 2021). Where authors have presented a potential mechanism, they often evoke lateral inhibition to some extent, a process whose involvement in ZT generation was first proposed by Norena et al. (2000). It should perhaps be noted that the following mechanisms need not be mutually exclusive.