A PREDICTIVE CODING PERSPECTIVE
ON SUGGESTION AND AUDITORY
FREQUENCY DISCRIMINATION
A literature review
Word count: 9228Andreas De Bleser
Student number: 01509904Supervisor: Prof. Dr. Durk Talsma
A dissertation submitted to Ghent University in partial fulfilment of the requirements for the degree of Bachelor/Master/Doctor of theoretical and experimental psychology
The empirical study that was originally planned, was unable to be conducted as a consequence of the COVID-19 pandemic. The concerning research question was how verbal suggestion could influence auditory frequency detection, in which the crucial manipulation was a strong expectation induced by the experimenter and the surrounding context of a research environment. The data collection was scheduled in the second half of March, right at the start of the corona pandemic, and was planned to be conducted at the anechoic room at the technologiepark in Zwijnaarde (operated by the acoustics workgroup). As such, data collection became impossible as online data collection was not an option. Therefore it was opted to write a theoretical literature review unifying different perspectives and identifying gaps in the literature concerning the influence of suggestion of pitch processing. This preamble was written in
theoretical and experimental psychology at University Ghent. I would like to express my deepest gratitude to the people who helped and supported me during the writing of this thesis in this relatively short time-frame and unforeseen circumstances.
First, I would like to thank Prof. Dr. Talsma for providing me with thoughtful discussions, practical help, and guidance in the writing process. He provided me with all the necessary tools to come to the completion of this project and supported me in aspects where it was most needed. I would also like to express my sincerest
appreciation towards Matthias Raemaekers and Isaiah Silverstein for the thorough proofreading of the draft versions and helping me in providing a clearer way to convey my message. Next, I would like to thank my friends from the courses Experimental and Theoretical psychology for the many interesting perspectives on a multitude of topics, helping me in developing a more critical way of thinking, and supporting me during my time as a student. To conclude, I am very thankful for all the opportunities I was given due to the hard work of my parents.
communication and music. The pitch perception literature is mostly focused on the relation between the physical properties of sound and the perceived pitch. However, in some cases perception is governed by strong, precise top-down expectations instead of bottom-up signals, as proposed by the predictive coding framework. The ascending and descending neural pathways which allow such interaction between central
processes and peripheral processing of pitch. Furthermore, are others able to influence our expectations, and consequently our perception, through a verbal suggestion. In this review it is argued that expectations are the result of inferences about the relation of prior expectations and verbals suggestions, which act as relational context cues, can influence these inferences. As a result of these verbally modulated expectations, are the forward and backward connections from and to cochlea influenced, which could result in a modulation in the discrimination of frequency congruent to the verbal suggestion.
Introduction ... 1
Auditory Perception and Pitch ... 1
Neurobiology and Neural Encoding of Pitch ... 4
Predictive Coding ... 8
The Hierarchical Organization of Pitch Perception ... 9
Precision Processing in Pitch Perception ... 12
Expectations and Verbal Suggestion ... 14
Thought Experiment: Suggestion Of Improved Frequency Perception ... 16
Discussion ... 19
Expectations can have a strong influence on perception. For example, when listening to a song that is played on the radio with poor radio reception, a familiar song will sound much clearer compared to an unfamiliar song. This is because we have an expectation of how the familiar song will sound. A historic example of such a song is "White Christmas" as recorded by Bing Crosby that was very popular in the 1960s. In a now-classic demonstration, people actually reported hearing this song when instructed to imagine hearing it (Barber & Calverley, 1964; Mintz & Alpert, 1972). This effect became better known as The White Christmas effect. Participants even report detecting the song embedded in a white noise tape even though the presence of the song was merely alluded in the instructions, but was, in reality, never presented (Merckelbach & van de Ven, 2001).
Classic models of perception have difficulties explaining this kind of effect. The fundamental goal of these models is linking sensory inputs to perceptual experience. Accordingly, perception is seen as a strictly hierarchical process with little to no influence of top-down processes. For example, a classic model of vision considers visual perception as a three-stage process of encoding, selection, and decoding. Visual inputs are encoded into neural signals, of which a small fraction are selected, resulting in a visual percept. Thus, these kinds of models only explain perception as indirect results of sensory input, that is, the content of our perception needs to be represented to some degree in the content of the signal we are trying to perceive. The idea that the content of perception is predominantly determined by the sensory signal, is the
foundation of many models of auditory perception. However, sometimes the content of our perception is the result of expectations and entirely unrelated to the content of the signal, as demonstrated by the aforementioned White Christmas effect. Such
phenomena can only be explained when top-down expectations are considered as an integral part of perception and if these expectations can be influenced by the
suggestion of others. The present review will focus on the influence of top-down expectations on the perception of pitch, how verbal suggestions are able to modulate these expectations, and to what extent external influences can be accounted for theoretically within the areas of pitch processing, predictive coding, and suggestion. This will be done by considering relevant studies in the field of physiology,
psychoacoustics, perception, learning psychology, and computational modeling. Auditory Perception And Pitch
Auditory perception provides us the ability to access the acoustic world around us. It allows us to detect, localize, and understand sound; abilities that are essential for
human communication. The abilities of our auditory system are astonishing considering that the sensory input is, in most cases, polluted by noise and only available for a very brief period in time. Our eardrums move to changes in air pressure, providing us with only two continuous measures on each side of our heads. This simple vibrating movement gives us access to a rich auditory percept and this perceptual wealth becomes even more impressive when we consider that a sound seldom occurs in isolation. The sound waves that reach our ears are a mixture composed of all the sounds occurring at that given moment, and in most cases, we are still able to hear someone talk to us in a bar with other people talking next to your table and music playing in the background (i.e, the cocktail party phenomenon; Cherry, 1953). From this mixture of sound waves, we are able to perceive differences in pitch, which helps us to select between sound sources and direct attention to the ones interesting to us.
In fact, each sound is the result of one or a combination of sound waves, and each sound wave consists of three defining physical properties: amplitude, frequency, and phase. Firstly, the amplitude is the maximum value of sound pressure caused by a sound wave and it defines the volume of a tone. Increasing amplitudes typically
correspond with non-linear increases in perceived loudness. At higher amplitudes (e.g., 100 dB above the hearing threshold), increasing amplitudes may not result in an
increased perceived loudness (Moore, 2013). Secondly, the frequency is the number of cycles in 1s of a sine wave and is the main characteristic that defines the perceived pitch of the sound (e.g., a high or a low tone). Finally, the phase of a sound wave is the fractional part of a period through which the sound wave has advanced. In other words, it is a measure of the time a sound wave takes to complete his cycle. Phase gains importance when a sound is the result of a complex sound wave. When two or more sound waves combine, the difference between the two phases of the waves
determines the resulting waveform. In other words, the phase difference describes the relation between different sound waves. Sound waves consisting of a single sinusoidal waveform with a single frequency are the simplest forms of sound and are known as pure tones.
However, there is not a linear relationship between the physical properties of the sound wave and the perception of these properties, as already briefly touched upon in the case of amplitude. The non-linear relationship makes sense when we consider that our auditory system has most likely evolved for the benefit of our survival and that it is, therefore, far from perfect. Instead, the auditory system prioritizes perception of the features that benefit us the most. In most cases, features that are necessary for
surviving our environment are prioritized for further processing. More generally, it can be argued that human perception only processes features of our environment that are behaviorally relevant and have some benefit in one way or another. In that case, it could be argued that the content of perception is far from an objective representation because it only processes relevant information. The lower level processes need top-down information in order to process the relevant information. As such, it could be proposed that in most cases, if not all, there are top-down influences involved in bottom-up processing.
Pitch perception, for example, is generally understood as the subjective process of assigning a frequency to a tone (ANSI, 2013) and the top-down influence on pitch perception becomes clear when these influences result in a modulated perception of the sound stimulus. In other words, when we experience an auditory illusion. Let us consider the case of illusionary pitch in a complex tone as an example. A complex tone is composed of a number of sinusoids or harmonics, each of which has a frequency that is an integer multiple of the frequency of a common (i.e., fundamental) component. For example, a note A3 played on the piano has a fundamental or first harmonic with a frequency of 220 Hz, the second harmonic with a frequency of 440 Hz, and so on. Interestingly, when the fundamental frequency is missing, the pitch of such a complex tone is perceived the same as if the fundamental was present. This is known as the missing fundamental effect (Licklider, 1954) and is widely studied in the field of pitch perception.
Other topics in the literature investigating the relationship between pitch and frequency are changes in pitch in function of the volume, the size of the frequency difference limen, and changes in difference limen in function of frequency, however, this review will focus mostly on the last two. The difference limen for frequency (DLF) is the smallest detectable change in frequency, which is the lowest for frequencies in the range of 1 - 2 kHz. In other words, humans show the greatest sensitivity to frequency changes in the range of 1 - 2 kHz. A computational model (Micheyl et. al., 2012) using pure-tone DLF data from 12 studies described the function between log-transformed normalized DLF’s and the stimulus parameter (i.e., frequency, duration, and level) as a linear combination of power functions. The authors concluded that humans are able to detect changes in frequencies that differ as little as 0.2%. However, the model only included data from DLFs for frequencies of 8 kHz or less. Moore and Ernst (2012) looked at DLF, expressed as a proportion of center frequency, on a much larger range (from 2 kHz up to 15 kHz).
Figure 1
Mean DLFs for the frequency range 2kHz - 15kHz
Note. Showing the mean values of DLF as a proportion of center frequency across participants at 20 dB (open circle) and 70 dB (open square) above the sound threshold. The mean of the two sound levels (filled square) shows a monotonic increase of DLF from 4 - 8 kHz and a plateau above 8 kHz. Data from Moore and Ernst (2012).
These results show a monotonic increase of DLF from 4 to 8 kHz, similar to the model of Micheyl et. al. (2012), however, at higher frequencies the DLFs seem to plateau (Fig. 1). In summary, the human auditory system is very sensitive towards differences in frequency, but this sensitivity decreases at the higher and lower end of the frequency spectrum. However, a question, which remains to be answered is how our auditory system is organized in order to show such abilities.
Neurobiology And Neural Encoding Of Pitch
Within the auditory system is the cochlea, which is located in the inner ear, the main organ responsible for our frequency selectivity and sensitivity. The incoming sinusoidal signal is propagated through the entire basilar membrane (situated in the cochlea). The peak in the pattern of vibration of the basilar membrane differs according to the frequency of the stimulus. High frequencies peak more towards the narrow and stiff base whereas low frequencies reach their peak more towards the wide and narrow apex. These properties of the basilar membrane are essential for the tonotopic
organization and in addition, is the separation of frequencies along the basilar
membrane supported by the outer hair cells within the cochlea. These outer hair cells help by sharp tuning the cochlea (Guinan, 2006, Fettiplace, 2020) and are the outer end of the olivocochlear function bundle (Fig. 2A).
This efferent system (more specifically, the medial olivocochlear system)is believed to be part of a reflexive brainstem control mechanism that mediates peripheral effects by supporting adaptation and frequency selectivity in the cochlea (Fig. 2B; Collet et. al.,1994; Maison et. al., 2001). However, central processes, such as attention, are thought to indirectly influence cochlear tuning via modulatory signals projected via this efferent system. For example, patients with impairments in this system have a decreased ability to selectively focus their attention on the frequency domain (Scharf et. al., 1997). Furthermore, selective attention is able to mediate the otoacoustic emission (OAE) (Walsh et. al., 2015; Beim et. al., 2018) and this OAE is assumed to be generated by the movement of the outer hair cells and is used as an objective measurement of frequency selectivity (Shera et. al., 2010). In other words, selective attention is able to sharp tune the cochlea.
Figure 2
Cross-section of the cochlea and the midbrain
Note: A, Cross-section of the inner ear. The frequency information arises from the inner ear cells, which are connected to the basilar membrane. B, The auditory brainstem. frequency information arises via the afferent pathway from the basilar membrane and the descending medial olivocochlear system projects to the outer hair cells. Adapted from May et. al. (2004).
How pitch is extracted from the acoustic wave in the cochlea and how
differences in pitch are represented in the auditory system is handled by two broad and distinct theoretical accounts: place-rate theories and temporal encoding theories. Generally speaking, place-rate theories state that pitch is extracted based on the tonotopic organization of the cochlea and accordingly, frequency discrimination (the ability to detect changes in frequency over time) and frequency selectivity are closely related. Consequently, place-rate theories claim that frequency discrimination is the result of changes in the excitation pattern of the auditory nerves exceeding a threshold value (e.g., ca. 1 dB, in line with amplitude modulation detection thresholds). In
contrast to this, temporal theories state that the spike timing information in the auditory nerve is used to extract frequency content due to the precise relationship between waveform and spike timing, as opposed to place information. Spikes occur at a given phase in the waveform, also known as phase locking. Accordingly, frequency
discrimination is the result of differences in phase-locking as suggested by temporal theories. Therefore, temporal theories assume that the relation between frequency selectivity and discrimination is not as close as assumed by the place theories.
As stated above, basic place theories assume that DLFs should be similar to cochlear tuning curves and the cochlea shows sharper tuning from apex to base (i.e., the cochlea is more sensitive to higher frequencies; Fig. 3A; Bentsen et. al., 2011).
Figure 3
Cochlear tuning and DLF
Note. A, means of cochlear tuning as measured by QERB. A larger QERB corresponds with a
larger OAE, in other words, a sharper cochlear tuning. Data from Betsen, Harte, and Dau (2011). B, comparison of DLFs with historical studies of cochlear tuning. B, DLFs as predicted by the model of Micheyl et. al., 2012 compared to measures of cochlear tuning provided by Glasberg and Moore (1990), and Oxenham and Shera (2003).
Conversely, DLFs deteriorate at the lower and higher ends of the frequency spectrum (i.e., <500 Hz, >4000 Hz; Fig. 3B), similar to the deterioration rate of phase-locking in the auditory nerve in small mammals (Heil & Peterson, 2015). The evidence favors temporal coding as the main mechanism of pitch encoding, since place coding
assumes a better ability in frequency discrimination at higher frequencies, whereas the opposite is found. However, the relative importance of temporal and place information has been heavily debated in hearing science and it is generally agreed upon that both types of encoding are necessary for pitch perception.
Phase locking is the most detailed method of pitch encoding and is effective up to a certain upper-frequency limit. At higher frequencies, the coarser place information is used instead. However, there is no consensus regarding the upper limit of phase locking in monaural processing, such as pitch extraction, as compared to the agreed-upon 1500 Hz upper limit (Burghera et. al., 2013) for binaural processing (e.g,
interaural time differences). Estimates of the upper limit of pitch extraction phase-locking vary from 1500 Hz to 10000 Hz (Verschooten et. al., 2019). It has been questioned how and why the upper limit of pitch extraction would be higher than the upper limit of binaural processing. Moore et. al. (2012) argued that the increasing DLFs up to 8000 Hz indicates that phase locking is effective and the plateau at higher
frequencies indicates place encoding (Fig. 1). If phase locking is not effective for frequencies above 1500 Hz, how do we explain the evidence for relatively good performance in frequency discrimination up to 4000-5000 Hz?
Oxenham (2018) suggests that this is the result of pitch perception bound by central, rather than peripheral encoding constraints. This idea is supported by the finding that frequency discrimination can be trained (Amitay et. al., 2005; Amitay et. al., 2006; Wright & Zhang, 2009), which could be interpreted as an influence of the rapid plasticity in central processes. The deterioration of DLFs at high frequencies might be due to the lack of exposure to very high frequencies in daily life (Lau et. al., 2017) and in addition, a computational modeling study (Micheyl et. al., 2013) showed that noise correlation is able to solve the large differences in frequency modulation and amplitude modulation sensitivity, using a place-rate code without the need of temporal encoding, which makes it possible for place theories to explain our ability to discriminate changes in frequencies. Consequently, it could be argued that place coding is able to explain human frequency discrimination on the condition that central processes are integrated with the bottom-up processing of pitch.
Predictive Coding
The notion that pitch perception is modulated by central processes fits with a recently proposed theoretical framework of perception that is known as predictive coding (Rao & Ballard, 1999; Friston, 2005; Friston, 2010; Clark, 2013). The central idea of this framework is that the brain is an active hypothesis testing organ, that makes sense of the world by making causal inferences about the expected state of our external environment. This idea can be traced back to Helmholtz (1866) and has been refined in the predictive coding framework using mathematical models grounded in Bayesian frameworks. Predictive coding provides a comprehensive framework about neural functioning and arguably the dominant theoretical framework to describe the wide range of domains in perception (Friston, 2010)
Predictive coding states that we make sense of our environment by making causal inferences about it, by creating predictions about what will happen next (Hohwy, 2012). These predictions are contextually dependent (Feldman & Friston, 2010) and hierarchically organized (Friston, 2008; Harrison et. al, 2011): lower lever ‘driver’ neurons receive and process the initial sensory information and send the regularities of the noisy world to the ‘integration’ neurons. The latter, in turn, creates an abstract interpretation of these regularities, which leads to the ‘creation’ of a model. This model contains the expectation of a future state, which is used to inform the lower level driver neuron by giving it a ‘best guess’ of what they will perceive. At the lower level, the difference between the prediction and actual sensory input elicits a prediction error. This prediction error is used as a signal to the higher-level neurons that the model needs to be updated.
However, not all prediction errors are created equally. Just like in standard hypothesis testing of mean differences, for example, in a Student’s t-test, means can be seemingly different but the variability in the data may prevent the conclusion that the means actually differ. In other words, whether the observed means actually differ depends critically on the precision of the underlying data. In terms of the predictive coding framework, the system will revise the internal model based on the first-order statistic (the predictions errors) and those errors are weighted by precision. Hence, like any other judgment of difference, the prediction must be weighted by the precision. A precise prediction error will adjust the model more strongly than an imprecise one. Just like the prediction itself, the precision is context-dependent, and thus, the expectation of precision needed. Accordingly, the model should contain both first-order prediction (the expectation itself) and the second-order prediction (the expectation of precision).
What does this mean in terms of perception? When the expected precision is high, the bottom-up sensory signal will have more weight and have a stronger influence on perception compared to the expectation. Whereas when the expected precision is low, the top-down expectation will have a relatively stronger influence on perception when processing a signal. This precision processing (i.e., balancing top-down and bottom-up influences) maps very well on what is generally described as the function of attention. As an example of this precision processing, say that we point a flashlight in a dark room, filled with objects. The object that is illumined by the searchlight will have a much higher expectation of precision compared to the objects in the dark. This is roughly equivalent to the spotlight (or searchlight) metaphor, which is often used to describe the functional role of attention, which maps very well on the function of precision processing in predictive coding. Mechanistically, attention is described as a procedure that weights or biases certain sensory channels, and on a neurobiological level this is believed to be accomplished by modulating the sensory gain (Martinez-Trujillo & Treue, 2004; Kok et. al., 2011; Kok et. al., 2012). Within the predictive coding framework, precision processing is assumed to have the same function, mechanism, and neurological implementation as attention (Hohwy, 2012).
The Hierarchical Organization Of Pitch Perception
The theoretical framework of predictive coding within the context of auditory perception can be a powerful tool to broaden our knowledge on how pitch is processed. If we assume that pitch processing is organized in a recurrent hierarchical fashion, then the top-down neurons in the auditory cortex are able to send information to the driver neurons of what they should expect to perceive. However, is it neurologically possible for the neurons in the cortex to interact with lower-level neurons responsible for the encoding of pitch?
The peripheral pitch encoding system, connecting the cochlea and the auditory cortex via ascending and descending pathways, provides strong evidence for this hierarchical organization with the possibility of interaction between higher-level integration neurons and lower-level driver neurons. More specifically, the major ascending pathways arise from the cochlea and form forward connections to
subcortical structures (among which, the cochlear nucleus, superior olive, and inferior colliculus), the thalamus (more specifically, the medial geniculate nucleus) and
continuing up to the auditory cortex (Fig. 4). The olivocochlear system is the major efferent pathway of pitch perception and descends from the superior olivary complex to the outer hair cells of the cochlea. In addition, animal studies suggest a wide range of
backward connections from the auditory cortex reaching almost all the parts of the ascending pathway and the olivocochlear system (see Winer, 2005 for a review). Similarly, A functional magnetic resonance imaging (fMRI) study (Yakunina, et. al., 2019) showed a strong relation in activation between the superior olivary complex and the auditory cortex during the presentation of a target pure tone. This suggests an efferent attention-related connection, which makes it possible for the auditory cortex to influence frequency selectivity in the cochlea as a result of sharper cochlear tuning (Walsh, et. at., 2015; Beim, et. al. 2018).
Figure 4
Overview of the afferent and efferent pathways of the auditory system
Note. The afferent pathway (black lines) connects the cochlea with the auditory cortex and the efferent pathway (blue lines) connect the auditory cortex with all parts of the afferent pathway. Adapted from Winer (2005)
The organization and localization of pitch within the auditory cortex is still under debate. There is consensus regarding an important role of the medial part of Heschl’s gyrus as the primary auditory cortex. Conversely, there is no consensus regarding the role of non-primary areas for pitch. Some fMRI studies have suggested a pitch center situated in the lateral part of Heschl’s gyrus (Patterson et. al., 2002; Hall et. al., 2006; Puschmann et. al., 2010). The lateral part of Heschl’s gyrus is the human homolog of a region of neurons sensitive to both pure tones and missing fundamentals with similar pitches found in Marmosets, primates with a similar hearing range as humans (Bendor & Wang, 2005).
However, it has been argued that pitch processing is distributed over multiple neural regions. Single-cell recordings of ferrets (Bizley et. al., 2013) and marmosets (Bendor et. al., 2010) show broadly distributed pitch responses in the auditory cortex. In humans, temporally regular sounds below 30 Hz are unable to produce a percept of pitch (Pressnitzer et. al., 2001) and by using this lower limit of pitch of temporally regular sounds, a dichotomy in pitch responses in the auditory cortex has been found. More specifically, evoked responses are found above and below the lower limit of pitch, whereas induced high gamma oscillations are only found at repetition rates in the range that produce pitch. This dichotomy makes the induced response a more viable option as a neural correlate of pitch processing (Griffiths et. al., 2010). Using this induced pitch response, activation along the entire gyrus of Heschl and the planum temporale, the region posterior to Heschl’s gyrus, was found using direct
electrophysiological recordings from eight patients suffering from epilepsy (Gander et. al., 2019). This suggests an important role of Heschl’s gyrus and planum temporale in pitch processing. In addition, a study using single-cell recordings and local field potentials (Kikuchi, et. al., 2019) in macaques showed a spatial separation between pitch sensitive neurons and gamma oscillations. Furthermore, the most sensitive pitch neurons in the cortex contained insufficient information to explain the sensitivity in frequency discrimination of macaques. Consequently, this suggests the need for a distributed cortical pitch processing system based on forward and backward
connections between regions in order to explain frequency discrimination. This corresponds with the results of the computational study of connectivity in Heschl’s gyrus (Kumar et. al. 2011) which shows that the lateral area of Heschl’s gyrus receives forward connection from, and sends backward connections to the medial and middle area. Suggesting a hierarchical organization of Heschl’s gyrus in which the lateral area
of Heschl’s gyrus is situated higher on the hierarchy than the medial and middle, which are situated at the similar, lower level of the hierarchy.
Precision Processing In Pitch Perception
The hierarchical organization of pitch processing described above allows the higher-level integration neurons to send a prediction of what pitch the lower-level driver neuron should expect. In return, the driver neuron is able to adjust the top-down pitch expectation if sensory information is not accurate. However, connectivity modulation between higher levels and lower levels of the pitch processing hierarchy, as a result of the aforementioned precision processing, should also be observed. This effect of precision processing can be observed within Heschl’s gyrus, as suggested by the modulation of connection strength between the lateral, medial, and middle part of Heschl’s gyrus resulting from temporal regularity (Kumar et. al., 2011). Increased temporal regularity leads to a stronger backward connection to, and a weaker forward connection from, the lower level medial and middle part of Heschl’s gyrus. These functional asymmetries (changes in connectivity strength) suggest that increased temporal regularity will produce an accurate prediction of the environment in the lateral part of Heschl’s gyrus. Prediction errors from the medial and middle part are
downweighted, and the higher-level model of the surrounding is upweighted. Similar to this, precision weighting makes it more likely for the top-down expectation of the environment to influence pitch perception, relative to the sensory signal.
Sharper cochlear tuning, as a result of precision weighting, is possible as suggested by the modulatory effects of selective attention on cochlear tuning in electroencephalography (EEG) studies (Dragicevic et. al., 2019) and OAE recordings (Walsh et. al., 2015; Beim et. al., 2018). The instructions of the task will result in an expectation and, as a result of guiding the attention of the participants, the expected precision of the prediction error will be low, which would result in an increased weight of the backward connection, which could result in a sharper tuning of the cochlea. By applying precision processing in pitch processing, it becomes apparent how top-down expectations are able to result in illusory percepts such as the missing fundamental. For example, a single harmonic is able to cause a missing fundamental percept in the presence of background noise, but the same single harmonic is not able to induce a missing fundamental effect in the absence of noise (Houtgast, 1976). In this study, participants were presented two successive complex tones, with different fundamental frequencies. The first tone contained six harmonics and the second tone contained initially three harmonics and was gradually decreased to one harmonic.
From a predictive coding perspective, trials where the second tone consisted of multiple tones would lead to a prior expectation. Trials where the second tone
consisted of just one harmonic, should elicit a strong prediction error. In turn, this would result in a revision of the model which predicts that the second tone consists of multiple harmonics. This is the case when the expected precision is high and the participant expects the signal to be reliable, in other words in the trials in which the noise is absent. However, in the presence of background noise, the expectation of precision is low, because the participant expects the signal to be unreliable. Consequently, is the signal, representing the sound of one harmonic, downweighted and the top-down expectation of hearing multiple harmonics upweighted. This precision weighting in a noisy context makes it more likely that the model selected for conscious perception would be influenced by the prior expectation, which would then result in the percept of a missing fundamental. However, the precision weighting during the absence of noise makes it more likely that the model selected for conscious perception would be influenced by the bottom-up signal, which would result in the percept of a single harmonic.
In addition, presenting harmonic components successively can also result in a missing fundamental percept (Hall & Peters, 1981; Grose et. al., 2002). In these tasks, no prior trials were used, which excludes the possibility that a prior expectation could be the result of hearing one coherent tone. In daily life, we, in the majority of the time, get exposed to coherent complex tones and only on rare occasions we encounter successive pure tones. Our daily exposure to coherent complex tones could result in the prior expectation of hearing these kinds of tones. In these tasks the background noise may have caused the expected precision to be low, which downweights the bottom-up signal and upweights the top-down expectation of hearing a coherent complex tone, which results in the percept of a missing fundamental. On the other hand, as the interval between the successive harmonics increases, the ability to perceive a missing fundamental decreases (Grose et. al., 2002). When the harmonics succeed each other fast, the sensory input and the expectation are not that different, so our prior expectation of hearing a coherent complex tone is fairly accurate. However, if the interval between the successive tones increases our model starts to be less
accurate and is unable to provide an explanation for why the prediction error is so big, which causes the model to be updated. The prediction of hearing a coherent complex sound is downweighted and the sensor signal is upweighted, making the perception of hearing separate harmonics more likely.
In short, the neural and functional asymmetries of the pitch perception system makes predictive coding a viable framework to explain the processing of pitch. By applying the predictive coding framework, is the importance of expectations and context illuminated as an intricate part of pitch processing. However, the perception of pitch is rarely a purely individual act since pitch perception is the fundament of a lot of auditory processes, such as the perception of music and speech, which in essence is the perception of sound made by others. As such, it is important to understand the influences of others on our perception as a result of influencing our expectations. This raises the questions to what extent -and how- others are able to influence these expectations.
Expectations And Verbal Suggestion
The effects of others on expectations by means of verbal suggestions have been investigated extensively in clinical contexts, in particular, in light of the placebo effect in pain research. Placebo is defined as a substance or procedure that has no inherent power to produce an effect that is sought or expected (Stewart-Williams & Podd, 2004). An fMRI study on the expectancies of nicotine in smokers showed activation in smoking-related areas when told to expect nicotine (Faria et. al., 2020), which suggests that verbal suggestions can create expectations. These verbally induced expectations are able to influence the outcome of selective serotonin reuptake inhibitor (SSRI) treatment in patients with social anxiety. The group who were told to expect great improvement showed an outcome that was two to three times greater than the group who were told not to expect improvement (Faria et. al., 2017). Verbal
induced expectations are also capable of modulating the effects of an objectively effective analgesic (Benedetti et. al., 2003). Understanding the underlying mechanisms of placebo and the placebo effect could provide insight on how verbal suggestions are able to modulate expectations.
Two different mechanisms of the placebo effect have been proposed. One the one hand, classical conditioning and on the other, the expectancy theory. According to the classical conditioning approach, the placebo effect is a conditioned response which results from the pairing of the placebo with an effective treatment. For example, pairing an inactive cream with a decreased pain sensation results in a strong decreased pain sensation when the cream is applied compared to trails where no cream is applied, even though the intensity of pain is the same (Voudouris et. al., 1990). Thus, the classical conditioning account states that the placebo effect is the result of a contingent relation between the suggestion and the treatment. In response to the classical
conditioning account, Kirsch (1985) developed the expectancy theory, which assumes that placebo effects are the result of conscious expectancies concerning the placebo intervention. These conscious expectancies could be the result created and modified by classical conditioning but also learned by observation or verbal instruction. Recent reviews (Bräscher et. al., 2017; Babel, 2019) conclude that the placebo effect is the result of conscious (e.g., learning by observation and verbal suggestion) and
unconscious learning (e.g., classical conditioning). However, new developments from learning psychology have shed new light on learning. The core idea is that in verbal organisms, even simple learning, such as classic conditioning, are instances of
complex behavior. Complex learning can be defined as changes in behavior due to the joint effect of multiple regularities (all states in the environment of the organism that entail more than the presence of a single stimulus or behavior at a single moment in time) in the environment (De Houwer, & Hughes, 2020).
From this novel perspective, both, conscious and unconscious learning can be considered as instances of arbitrarily applicable relational responding (De Houwer & Hughes, 2017). Subsequently, there is only one underlying mechanism of placebo, namely arbitrarily applicable relational responding. A complete overview of arbitrarily applicable relational responses is well beyond the scope of this article but put in simple terms, it entails that regularities can influence behavior because regularities function as a relational contextual cue. In other words, relational contextual cues are moderators of learning. Looking at verbally induced placebo as an arbitrarily applicable relational response provides an interesting perspective on how verbal suggestions, describing a regularity, are able to induce changes in behavior. Let us reconsider the example of the placebo cream. Participants will report less pain when a cream is applied if they are suggested that a cream will decrease pain, even though the cream and the pain were never paired with each other. It is unlikely that this decreased pain reaction is caused by the (psycho)physical properties of the words ‘decreased pain’ and ‘cream’ occurring together in space and time, but rather that the suggestion functioned as a relational context cue. The suggestion makes people respond as if the cream is a predictor of decreased pain. Purely verbal placebos are able to elicit a placebo effect because the verbal suggestion functions as a relational contextual cue of equivalence.
If behavior relies on the relations between stimuli on a cognitive level, then the organism needs to at least have propositional knowledge about the nature of the relation. Propositions are informational units that specify elements, but also how those elements are related (De Houwer et. al., 2016). A core assumption of propositional
models is that propositions can be combined via inferential reasoning (i.e., new knowledge can be derived from existing knowledge). If making new inferences forms the base of arbitrarily applicable relational responses, then one should be able to influence these responses by influencing the inferences, for example, by providing information that influences the probability of the formation of certain inferences (De Houwer et. al., 2020). Expectations can be the result of implicit knowledge about statistical regularities, which is then used incidentally to adapt behavior, or they can stem from explicit beliefs about specific events, others, and objects. In other words, it can be argued that expectations are propositions. Let us reconsider the
aforementioned example of the placebo cream again. Verbal suggestion (relational contextual cues) leads to decreased pain sensation because a derived proposition is created that cream is related to decreased pain. In other words, cream is expected to result in decreased pain. The inference resulting in the specific expectation that the cream results in decreased pain, is the outcome of the verbal suggestion increasing the probability of this specific inference.
The underlying mechanisms of pain placebo support the idea that expectations are able to modulate perception. These conscious and unconscious expectations are the result of inferences about the relations between prior expectations. Multiple inferences, and consequently multiple resulting expectations, are possible. The inference resulting in the posterior expectation is guided by the verbal suggestion and the suggestion is able to do so by increasing the probabilities of a given inference. Similarly, the expectations in pitch perception are influenced by verbal suggestions because they act as evidence for a certain inference and therefore increase the probability of a given expectation.
Thought Experiment: Suggestion Of Improved Frequency Perception In the following thought experiment, we will integrate our newly acquired knowledge and make a case that a verbal suggestion is able to modulate our predictions and accordingly modulate frequency perception as measured by a three interval alternative forced-choice task (3I-AFC), more specifically an AXB task. In this task participants are presented with three sequential tones, respectively the A, X, and B tone. The X tone (i.e., the reference tone) has a frequency F which remains constant during the experiment. The test tone (either the A or B tone) has a frequency of F + ΔF, where ΔF represents the frequency difference. The remaining tone has the same frequency as the reference tone (Fig 5.). Participants are instructed to indicate which tone differs from the others (i.e., the variable tone). After two consecutive correct
responses, the frequency difference (ΔF) is halved. After an incorrect response, ΔF is multiplied by the same factor. Crucially, half of the participants (test group) are led to believe that they are participating in the final step of product development of software that aims to optimize the users’ sound experience at home. These participants are told that the software objectively works, but that the product developers would like to know to what extent a person's experience is affected by the software. In this task, the software will enhance the frequency range of frequencies tested. However, the other half of the participants (control group) will not be informed about the enhancement of the frequencies. Actually, no enhancing software of any kind will be used, yet the test group will show an increased ability to detect differences in frequency, which ultimately will result in a decreased detection threshold compared to the control group.
The verbal suggestion that frequencies are enhanced will function as a relational contextual cue. The suggestion as an event will indicate that software enhancing frequencies are similar to other stimuli related to better perceiving
differences in frequency. Consequently, we are able to generate a more precise and accurate model about the environment due to the verbal suggestion. At the start of the experiment, participants will make a causal inference about the world by making a prediction of what will happen.
Figure 5
Example trial structure
The initial inference leading up to the expectation, is guided by the verbal suggestion that functions as context. The expectation of hearing differences in frequencies results in a modulation of the cochlear frequency tuning, corresponding with a modulated OAE by selective attention (Beim et. al, 2015; Walsh et. al., 2018), which increases the probability of detecting frequency differences. Consequently, there will be little difference between the expectation and the sensory input, in turn, resulting in a small prediction error. Furthermore, the broader context (product development, objectively working software, etc.) makes it more likely that not hearing a difference in frequency (i.e., a prediction error) is the result of noise and not due to an incorrect expectation. In other words, the expected precision of the prediction error is low (Fig. 6a).
Figure 6
Schematic of the assumed precision weighting
Note. Schematic of the assumed precision weighting. A, the initial weighting of the test group. The suggestion results in an increased weight of the expectation and a decreased weight of the bottom-up signal. B, the initial weighting of the control group. The control group will have a fairly inaccurate expectation of what they will perceive which will result in a decreased weight of the expectation and an increased weight of the bottom-up signal.
Detecting the frequency difference correctly would result in an increased weight of the backward connection from the auditory cortex to the cochlea due to the increased accuracy of the model. In turn, the frequency tuning is enhanced even more, which increases the probability of detecting the differences in frequency on the next trial. Inevitably, at some point in the experiment, the frequency difference is too small for the participant to detect and, therefore, elicits a prediction error. Initially, the prediction error is downweighted and consequently, the model will not be updated. After multiple incorrect responses, the accuracy of the expectation decreases, and the model is no longer able to explain the prediction error. Subsequently, the weight from the backward connection decreases, and the weight from the forward connection increases, resulting in an optimized model containing the participants' frequency discrimination threshold.
The model of the control group is less accurate and precise, consequently, a less accurate and precise expectation is made (Fig. 6b). This would result in a less strong backward connection to the cochlea. Therefore, we expect the cochlea to be not as sharply tuned, resulting in a decreased probability of detecting frequency
differences. The control group would, therefore, need more exposure to the differences (more training), compared to the suggestion group, in order to update their model for it to reach the same level of accuracy and precision as the initial model of the
participants who received the suggestion. Overall, this would likely result in a
diminished ability to detect frequencies and results in a higher detection threshold for the control group compared to the test group.
Discussion
In the present review, it was proposed that in order to explain the findings of pitch encoding literature (e.g., the good performance of frequency discrimination up to 4000-5000 Hz) there is a need to integrate central processing with peripheral
processing. Predictive coding provides an elegant solution that allows the integration of higher-level processes with lower-level perception by claiming that perception is the result of a recurrent hierarchical model of our environment. This model contains an expectation about what we will perceive and the expected precision of what we are about to perceive and is revised due to precision weighted prediction errors. When the expected precision is low, perception will be dominated by the prior expectations. The neural afferent and efferent pathways connect the cochlea, subcortical, thalamic nuclei, and the auditory cortex with each other. These pathways make the interaction between central and peripheral processes possible. The organization of pitch processing in the cortex is an ongoing topic of debate, however, evidence suggests a distributed pitch
processing system consisting of recurrent connections between higher and lower levels. The connectivity strength of these backward and forward connections is modulated in function of the expected precision, as was shown in computational, electrophysical, and behavioral studies. New perspectives on learning and inference provide evidence that learning is due to the joint effect of multiple regularities. The result of this joint effect is arbitrarily determined and, on a cognitive level, are the underlying expectations the result of probabilistic inferences. A given verbal suggestion is able to weight an inference in the direction of a given expectation by influencing the probabilities of the inferences. As such, the verbal suggestion of increased ability of frequency discrimination can influence central expectations, which in turn influences peripheral pitch processing, resulting in a sharper frequency tuning of the suggested frequency range. If the signal is incongruent to the expectation, then the cochlea could be tuned to a wrong frequency range and result in decreased ability to detect a
difference in frequencies. If the signal is congruent to the expectation, then the cochlea would be sharp tuned to the correct frequency range, which could result in a lower threshold for frequency discrimination.
The proposed effect due to verbal suggestions corresponds with the more general cognitive effects found in placebo, neuromodulation, and neurofeedback literature, which could all be interpreted as expectancy effects. Perception of pain can be decreased by a verbal suggestion of decreased pain (placebo analgesia), but pain perception can also be enhanced as a result of a suggestion of enhanced pain (nocebo hyperalgesia) (e.g., Benedetti et. al. , 2003; Colloca et. al., 2008). Studies using
transcranial direct current stimulation (tDCS), a non-invasive neuromodulatory device which hyperpolarizes and hypopolarizes neurons as a result of anodal and cathodal stimulation respectively (Stagg & Nitsche, 2011), suggest a performance enhancement of all kinds of cognitive tasks (e.g., visual change detection; Tseng et. al., 2012). However, a recent review (Horvath et. al., 2015) proposed that there appears to be no reliable cognitive effects of tDCS in a healthy population (for a commentary on these results read, Price & Hamilton, 2015). The null result of tDCS on a range of cognitive tasks could potentially be explained by differences in induced expectancies as a result of differences in the tDCS procedure. Similarly, neurofeedback (techniques that propel participants to self-regulated neural activity) is, supposedly, able to enhance perceptual sensitivity (e.g., lower visual detection thresholds; Scharnowski et. al. 2012). However, it has been suggested that neurofeedback does not have additional efficacy over placebo feedback (Schabus et. al., 2017; Thibault et. al. 2017), thus the enhanced
perceptual sensitivity could be the result of expectations induced by the neurofeedback procedure.
By considering expectations as an integral part of peripheral pitch processing, the dual mechanism of pitch encoding, involving place and temporal code, loses explanatory power. The deterioration of DLFs at the lower (<500 Hz) and higher ends (>4000 Hz) of the frequency spectrum can be explained because we are less exposed to these frequencies in daily life. For example, the highest note (C8) on the piano has a
frequency of about 4186 Hz. Less exposure to those frequencies results in a less accurate prior model, resulting in a decreased ability to perceive differences in these frequency ranges. Thus, the deterioration is a result of expectations and not of a switch for temporal to place coding. This is supported by the increasing need of place coding as processing moves up on the hierarchy due to the increasing coarseness of temporal information. For example, phase locking in the inferior colliculus is not observed above 1000 Hz (Liu et. al., 2006) and in the auditory cortex, phase-locking is only observed for frequencies up to 100 Hz (Lu & wang, 2000). Consequently, if temporal information is used to encode pitch in the auditory nerve then it needs to be transformed into place information at higher stages of processing.Contrary to temporal information can place information be maintained from the auditory nerve up to Heschl’s gyrus (Moerel et. al, 2014). This suggests place encoding as the defining mechanism of bottom-up pitch processing and it provides a rather provocative answer on the debate about the upper-limit of phase locking (Verschooten et. al, 2019), more specifically, that temporal information is not a necessary mechanism in the encoding of pitch as suggested by Oxenham (2018).
Most studies looking from a predictive coding perspective at auditory perception are agnostic about the meaning of expectations (Denham & Winkler, 2020), and based on the modeling literature, it is unclear whether expectations are the result of explicit interferences or merely due to implicit co-occurrence of variables. In addition, it is questioned what information (e.g., the timing of a stimulus) is conveyed by these expectations. In accordance with learning literature, this review proposes that expectations are the result of implicit inferences about relations between previously learned expectations (De Houwer et. al., 2020). These prior expectations contain information about the stimulus, but also how the stimulus is related to other stimuli (De Houwer, et. al, 2016). Consequently, the information conveyed by expectations
depends on prior expectations about stimuli and the relations between stimuli, and the inferences on these expectations. This review focused on verbal suggestions as a
method to influence these inferences by acting as a relational contextual cue of equivalence, making the paired expectations equal (if A then B is paired, by
suggestion, with B then C results in the equality A then C). However, depending on the context, relational contextual cues can function as opposition (if A then not C; e.g., Hughes et. al., 2019). Future studies investigating how, and when contextual cues change function could extend knowledge about what information is conveyed by prior expectations. In addition, there is a need for a broader understanding of context as a modulator in the weighting of prediction errors.
Integrating complex learning and PC could provide an interesting
conceptualization of individual differences in perception. Individual differences could influence perception in four distinct steps. Firstly, the learning history of an individual is decisive for the formation of relations and what prior expectations we have. Differences in learning history could result in differences in prior expectations. Second, the
probabilistic inference could be influenced by personality traits. For example, an anxious person could bias their inferences towards the expectations that stimuli elicit a fearful reaction. Third, the weight of contextual cues on these inferences could vary between individuals. For example, a highly suggestible participant could give more weight to verbal suggestions as contextual cues, which could result in an increased amount of false precepts. Finally, there could be differences in the weighting of
precision. For example, autism could be the result of a uniform and rigid high expected precision (Van de Cruys et. al., 2014) and tinnitus could reflect an excessive expected precision of the prediction error (Sedley et. al., 2016).
Chronic tinnitus is a common complaint defined as chronic ringing in the ears or head causing a severe impact on the individual’s health and wellbeing. Despite the prevalence and impact of tinnitus, there is as of yet no cure or consensus regarding the underlying mechanisms. However, a predictive coding model of tinnitus considers it as dysfunctional precision processing which results in a wrongful upweighting of the prediction error elicited by spontaneous noise in the auditory system (Sedley et. al., 2016). Modulating the expectancies of pitch, by the means of verbal suggestion, could provide important practical implications for the treatment of tinnitus. Evidence of effective treatment for tinnitus, such as frequency discrimination training, tDCS, and neurofeedback modulation could be interpreted as changes in the expectation of hearing tinnitus. Accordingly, the increasing accuracy of expectations normalizes the excessive precision weighting, resulting in a decrease in the symptomatic perception of tinnitus. Auditory frequency discrimination training has been found to have a positive
effect on tinnitus severity (Flor et. al., 2004). The effects of the treatment were best predicted by the days of training (a decreased frequency discrimination threshold). A meta-analysis on tDCS in tinnitus patients provides evidence for a reduction of 13.5% in tinnitus intensity (Song et. al., 2012). Similarly, chronic tinnitus patients show reduced loudness of tinnitus as a result of neurofeedback (Dohrmann et. al., 2007). Furthermore, the individual differences in the modulation of expectancies and precision weighting could explain the differences in the efficacy of the treatment and how, in some cases, the treatment results in an increase of tinnitus severity. Verbal
suggestions used in combination with existing treatments could change the expectation of pitch more effectively and increase the efficacy of the treatments, or simply by itself, the verbal suggestion could potentially result in a decrease of tinnitus severity.
The modulation of pitch also provides important implications for the perception of speech. In order to be able to understand what someone else is saying, we need to be able to perceive the sounds made by them. As such, pitch is also an important part of speech perception, more specifically is pitch important because it acts as a prosodic cue. In linguistics, prosody is concerned with those elements in speech that are not vowels and constants (e.g., intonation and stress). Consequently, understanding how pitch can get influenced could potentially lead to a better understanding of prosody in general. In tonal languages (e.g., Mandarin), shifts in pitch changes can alter the
meaning of a word, while in non-tonal languages (e.g., English) it is able to change only the meaning of a sentence. For example, making a statement versus asking a
question. If verbally modulated expectations can modulate pitch perception, then it is also very likely that others are able to modulate intonational pitch perception and therefore influence the meaning of utterances. In addition, it could be argued that the utterances of others in a reciprocal conversation, can modulate our expectations of what we will understand similar to a verbal suggestion. Given the ambiguity in speech, this could be a potential mechanic as to how we are able to understand each other. Another field of study in which the modulation of pitch could have practical implications in the field of music. Pitch is one of the main dimensions along which music varies and is the essential characteristic of melody. In addition, many accounts of music
perception claim that expectations and violations of the expectancies are intrinsic to music appreciation (Rohrmeier & Koelsch, 2012). Accordingly, it is possible for verbal suggestions to affect music appreciation. For example, some people will invest a lot of time and money in order to create the best sounding audio home system and will upgrade a component of their system because they heard a lot of good things about it.
After the upgrade, they report hearing details in music better even if the new
component objectively does not influence music. The expectation induced by reports of others, and perhaps smart, misleading marketing, could have enhanced the music experience. This example touches on the ethical questions that modulation of expectations due to verbal suggestion brings up. Is providing misleading information justified by enhanced musical experiences, increased therapeutic outcome, or
enhanced pitch perception? However, suggestions could also lead to increased pain, decreased, or worse therapeutic outcomes. To what extent is it ethically justified to interfere with the perceptual of others? In addition, if others are able to modulate our perception, then to what extent are we able to act freely on these modulated
perceptions. These questions are widely investigated in the philosophical free will debate, which is beyond the scope of the review (for a critical review of the role of neuroscience in the free will debate; Brass et. al., 2019).
In conclusion, the perception of pitch could be much more influenced by cognitive processes than previously assumed. As outlined by the predictive coding perspective is the sound signal only a necessary, but not sufficient condition to explain perception. The processing of this signal is influenced by cognitive processes
(expectations and expected precision). As such is the interaction between audiology, as the study of the sound signal, and psychology, as the study of cognitive processes, a necessity if we want a comprehensive knowledge of pitch perception. The cognitive processes of an individual are influenced by its learning history and the environment. Consequently, the influence of others, the background context, and the learning history should be an integral part of perception. These influences are often neglected in
perception literature, but insights of complex learning allow for a systematic
investigation of the interaction between these external factors. The remaining question is to what extent, and how these external factors influence can account for changes in perception.
References
Amitay, S., Hawkey, D. J., & Moore, D. R. (2005). Auditory frequency discrimination learning is affected by stimulus variability. Perception & psychophysics, 67(4), 691-698.
Amitay, S., Irwin, A., & Moore, D. R. (2006). Discrimination learning induced by training with identical stimuli. Nature neuroscience, 9(11), 1446-1448.
ANSI (Am. Nat. Stand. Inst.). 2013. American National Standard: acoustical terminology. Rep. S1.1-2013, Am. Nat. Stand. Inst./Accredit. Stand. Comm. Acoust., Acoust. Soc. Am., Washington, DC/Melville, NY
Bąbel, P. (2019). Classical conditioning as a distinct mechanism of placebo effects. Frontiers in Psychiatry, 10, 449.
Barber, T. X., & Calverley, D. S. (1964). An experimental study of" hypnotic"(auditory and visual) hallucinations. The Journal of Abnormal and Social Psychology, 68(1), 13.
Beim, J. A., Oxenham, A. J., & Wojtczak, M. (2018). Examining replicability of an otoacoustic measure of cochlear function during selective attention. The Journal of the Acoustical Society of America, 144(5), 2882-2895.
Bendor, D., & Wang, X. (2005). The neuronal representation of pitch in primate auditory cortex. Nature, 436(7054), 1161-1165.
Benedetti, F., Pollo, A., Lopiano, L., Lanotte, M., Vighetti, S., & Rainero, I. (2003). Conscious
expectation and unconscious conditioning in analgesic, motor, and hormonal placebo/nocebo responses. Journal of Neuroscience, 23(10), 4315-4323.
Benedetti, F., Pollo, A., Lopiano, L., Lanotte, M., Vighetti, S., & Rainero, I. (2003). Conscious
expectation and unconscious conditioning in analgesic, motor, and hormonal placebo/nocebo responses. Journal of Neuroscience, 23(10), 4315-4323.
Bentsen, T., Harte, J. M., & Dau, T. (2011). Human cochlear tuning estimates from stimulus-frequency otoacoustic emissions. The Journal of the Acoustical Society of America, 129(6), 3797-3807. Bizley, J. K., Walker, K. M., Nodal, F. R., King, A. J., & Schnupp, J. W. (2013). Auditory cortex
represents both pitch judgments and the corresponding acoustic cues. Current Biology, 23(7), 620-625.
Bräscher, A. K., Witthöft, M., & Becker, S. (2018). The underestimated significance of conditioning in placebo hypoalgesia and nocebo hyperalgesia. Pain Research and Management, 2018. Brass, M., Furstenberg, A., & Mele, A. R. (2019). Why neuroscience does not disprove free
will. Neuroscience & Biobehavioral Reviews.
Brughera, A., Dunai, L., & Hartmann, W. M. (2013). Human interaural time difference thresholds for sine tones: The high-frequency limit. The Journal of the Acoustical Society of America, 133(5), 2839-2855.
Cherry, E. C. (1953). Some experiments on the recognition of speech, with one and with two ears. The Journal of the acoustical society of America, 25(5), 975-979.
Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and brain sciences, 36(3), 181-204.
Collet, L., Bouchet, P., & Pernier, J. (1994). Auditory selective attention in the human cochlea. Brain research, 633(1-2), 353-356.
Colloca, L., Sigaudo, M., & Benedetti, F. (2008). The role of learning in nocebo and placebo effects. Pain, 136(1-2), 211-218.
De Houwer, J., & Hughes, S. (2017). Environmental regularities as a concept for carving up the realm of learning research: Implications for Relational Frame Theory. Journal of contextual behavioral science, 6(3), 343-346.
De Houwer, J., & Hughes, S. J. (2020). Learning to like or dislike: Revealing similarities and differences between evaluative learning effects. Current Directions in Psychological Science.
De Houwer, J., Hughes, S., & Barnes-Holmes, D. (2016). Associative learning as higher order cognition: Learning in human and nonhuman animals from the perspective of propositional theories and relational frame theory. Journal of Comparative Psychology, 130(3), 215.
De Lange, F. P., Kok, P., Rahnev, D., Jehee, J. F., & Lau, H. C. (2011). Attention reverses the effect of prediction in silencing sensory signals. i-Perception, 2(4), 341-341.
Denham, S. L., & Winkler, I. (2020). Predictive coding in auditory perception: challenges and unresolved questions. European Journal of Neuroscience, 51(5), 1151-1160.
Dohrmann, K., Weisz, N., Schlee, W., Hartmann, T., & Elbert, T. (2007). Neurofeedback for treating tinnitus. Progress in brain research, 166, 473-554.
Dragicevic, C. D., Marcenaro, B., Navarrete, M., Robles, L., & Delano, P. H. (2019). Oscillatory infrasonic modulation of the cochlear amplifier by selective attention. PloS one, 14(1). Faria, V., Gingnell, M., Hoppe, J. M., Hjorth, O., Alaie, I., Frick, A., ... & Carlbring, P. (2017). Do you
believe it? Verbal suggestions influence the clinical and neural effects of escitalopram in social anxiety disorder: a randomized trial. EBioMedicine, 24, 179-188.
Faria, V., Han, P., Joshi, A., Enck, P., & Hummel, T. (2020). Verbal suggestions of nicotine content modulate ventral tegmental neural activity during the presentation of a nicotine-free odor in cigarette smokers. European Neuropsychopharmacology, 31, 100-108.
Feldman, H., & Friston, K. (2010). Attention, uncertainty, and free-energy. Frontiers in human neuroscience, 4, 215.
Fettiplace, R. (2020). Diverse Mechanisms of Sound Frequency Discrimination in the Vertebrate Cochlea. Trends in Neurosciences.
Flor, H., Hoffmann, D., Struve, M., & Diesch, E. (2004). Auditory discrimination training for the treatment of tinnitus. Applied Psychophysiology and Biofeedback, 29(2), 113-120.
Friston, K. (2005). A theory of cortical responses. Philosophical transactions of the Royal Society B: Biological sciences, 360(1456), 815-836.
Friston, K. (2008). Hierarchical models in the brain. PLoS computational biology, 4(11). Friston, K. (2010). The free-energy principle: a unified brain theory?. Nature reviews
Gander, P. E., Kumar, S., Sedley, W., Nourski, K. V., Oya, H., Kovach, C. K., ... & Griffiths, T. D. (2019). Direct electrophysiological mapping of human pitch-related processing in auditory
cortex. NeuroImage, 202, 116076.
Glasberg, B. R., & Moore, B. C. (1990). Derivation of auditory filter shapes from notched-noise data. Hearing research, 47(1-2), 103-138.
Griffiths, T. D., Kumar, S., Sedley, W., Nourski, K. V., Kawasaki, H., Oya, H., ... & Howard, M. A. (2010). Direct recordings of pitch responses from human auditory cortex. Current Biology, 20(12), 1128-1132.
Grose, J. H., Hall III, J. W., & Buss, E. (2002). Virtual pitch integration for asynchronous harmonics. The Journal of the Acoustical Society of America, 112(6), 2956-2961.
Guinan Jr, J. J. (2006). Olivocochlear efferents: anatomy, physiology, function, and the measurement of efferent effects in humans. Ear and hearing, 27(6), 589-607.
Hall III, J. W., & Peters, R. W. (1981). Pitch for nonsimultaneous successive harmonics in quiet and noise. The Journal of the Acoustical Society of America, 69(2), 509-513.
Hall, D. A., Edmondson‐Jones, A. M., & Fridriksson, J. (2006). Periodicity and frequency coding in human auditory cortex. European Journal of Neuroscience, 24(12), 3601-3610.
Harrison, L., Bestmann, S., Rosa, M. J., Penny, W., & Green, G. G. (2011). Time scales of
representation in the human brain: Weighing past information to predict future events. Frontiers in human neuroscience, 5, 37.
Heil, P., & Peterson, A. J. (2015). Basic response properties of auditory nerve fibers: a review. Cell and tissue research, 361(1), 129-158.
Helmholtz, H. V. (1866). Concerning the perceptions in general. Treatise on physiological optics,. Hohwy, J. (2012). Attention and conscious perception in the hypothesis testing brain. Frontiers in
psychology, 3, 96.
Horvath, J. C., Forte, J. D., & Carter, O. (2015). Quantitative review finds no evidence of cognitive effects in healthy populations from single-session transcranial direct current stimulation (tDCS). Brain stimulation, 8(3), 535-550.
Houtgast, T. (1976). Subharmonic pitches of a pure tone at low S/N ratio. The Journal of the Acoustical Society of America, 60(2), 405-409.
Hughes, S., Ye, Y., & De Houwer, J. (2019). Evaluative conditioning effects are modulated by the nature of contextual pairings. Cognition and Emotion, 33(5), 871-884.
Kikuchi, Y., Kumar, S., Baumann, S., Overath, T., Gander, P. E., Sedley, W., ... & Griffiths, T. D. (2019). The distribution and nature of responses to broadband sounds associated with pitch in the macaque auditory cortex. cortex, 120, 340-352.
Kirsch, I. (1985). Response expectancy as a determinant of experience and behavior. American Psychologist, 40(11), 1189.
Kok, P., Jehee, J. F., & De Lange, F. P. (2012). Less is more: expectation sharpens representations in the primary visual cortex. Neuron, 75(2), 265-270.
