Ten adults ranging in age from 19–37 years (mean = 28.5, SD = 5.5) were paid to participate in the study. All participants passed a hearing screening at 20 dB HL or better at 500, 1,000, 2,000, and 4,000 Hz in the left and right ears and had no reported history of neurological or otologic disorders. Procedures were approved by the Institutional Review Board of the Albert Einstein College of Medicine (Bronx, NY) where the study was conducted. The examiner described the procedure to all participants in accordance with the Declaration of Helsinki who subsequently gave written consent and were paid for their participation.

Results of a power analysis conducted using Statistica software (Tibco), with a two-tailed t-test for dependent means, a medium effect size (d = 0.50) and an alpha of 0.05. determined that a sample size of nine participants would yield power of 0.90 to detect differences. The total of ten participants included in the current study thus exceeds the number required to obtain sufficient statistical power.

Stimuli were naturally produced complex sounds (32-bit stereo; 44,100 Hz digitization) obtained from free online libraries (Appendix I). Three categories of sounds were presented, of which there were 25 different tokens of spoken speech, 25 different tokens of musical instruments, and 57 different tokens of various environmental sounds. Speech sounds were naturally spoken words (e.g., “hello,” “goodbye”); music sounds were taken from various musical instruments (e.g., piano, flute, bass); and environmental sounds were taken from a range of sources, including nature (e.g., water dripping), vehicles (e.g., engine revving), household (e.g., phone ring), and animals (e.g., bird chirp). We modified all the sounds to be 500 ms in duration, with an envelope rise and fall times of 7.5 ms at onset and offset using Adobe Audition software (Adobe Systems, San Jose, CA). To verify that 500 ms in sound duration was sufficient to identify and distinguish the sound categories (e.g., spoken word, instrumental, environmental), three lab members (who were not included in the study) categorized a set of 150 sounds. The final set of 107 sounds used in the study had unanimous agreement as belonging to a category of speech, music, or environment. All 107 stimuli were equated for loudness using the root mean square (RMS) amplitude with Adobe Audition software. Categorical sounds were calibrated with a sound pressure level meter in free field (Brüel and Kajaer, Denmark) and presented through speakers at 65 dB SPL with a stimulus onset asynchrony (SOA) of 1.1 s.

Participants sat in a comfortable chair in an electrically shielded and sound-controlled booth (IAC Acoustics, Bronx, NY). Stimuli were presented via two speakers placed approximately 1.5 m, 45° to the left of center and 1.5 m, 45° to the right of center from the seated listener.

There were two task conditions: passive auditory and active auditory. During the passive auditory condition, participants had no specific task with the sounds. They watched a captioned silent movie of their choosing during the presentation of the sounds. The experimenter monitored the EEG to ensure that participants were reading the closed captions. In the active auditory condition, participants listened to the sounds and performed a three-alternative forced-choice task. Participants were instructed to listen to and classify each sequential sound by pressing one of three buttons labeled on a response keypad that uniquely corresponded to the sound category (speech, music, or environment). Participants were not provided with any information about the patterned structure of the stimulus sequence at any time. Thus, the patterned structure could be extracted by implicit learning regardless of the condition in which they were presented.

A total of 3,840 stimuli were presented in 16 separately randomized stimulus blocks (240 stimuli per block), eight stimulus blocks per condition. Stimuli were presented in continuous sequences of 240 stimuli, patterned by categorical groups of four stimuli (spoken words, musical instruments, environmental sounds), with an equal distribution of the categories in each condition (0.33 speech, 0.33 music, and 0.33 environmental). Category switches and repeats occurred randomly within each stimulus block. There were no repeated stimuli within any of the stimulus groups. Every sound token was unique in each categorical group (e.g., the sounds of the instruments harp, piano, clarinet, and guitar could be repetitions in the category of the music group, Figure 1). Categories switched randomly after four stimuli 70% of the time overall (336 switch trials per condition), and randomly repeated categories after four stimuli 30% of the time (144 repeat trials per condition). Presentation of sound groups was quasi-randomized such that categories could only repeat one time. Thus, sounds occurred in groups of either four or eight repetitions of any category. Participants were not informed about the structure of the sequences at any time and there was no demarcation to indicate when the category switched, or when the category repeated within a stimulus block; sound tokens were presented isochronously throughout every stimulus block. Thus, position #1 stimuli were only a ‘first position’ stimulus based only with implicit detection of the patterned categorical grouping.

Task conditions were randomized across participants, with half of participants presented with the passive condition first and half presented with the active condition first. Recording time was approximately 35 min per condition, with a snack break at the halfway point at which time the participant was unhooked from the amplifiers and could walk around. Total session time including cap placement, recording time, and breaks was approximately 2 h.

A 32-channel electrode cap incorporating a subset of the International 10–20 system was used to obtain EEG recordings. Additional electrodes were placed over the left and right mastoids (LM and RM, respectively). An external electrode placed at the tip of the nose was used as the reference electrode. Horizontal electro-oculogram (EOG) was monitored with the F7 and F8 electrode sites and vertical EOG was monitored using a bipolar configuration between FP1 and an external electrode placed below the left eye. Impedances were kept below 5 kΩ at all electrodes throughout the recording session. The EEG and EOG were digitized (Neuroscan Synamps amplifier, Compumedics Corp., Texas, United States) at a sampling rate of 1,000 Hz (0.05–200 Hz bandpass). EEG was filtered off-line with a lowpass of 30 Hz (zero phase shift, 24 dB rolloff).

This report includes data from all 10 participants in the study. There were no exclusions.

Behavioral Data: Hit rate (HR) and reaction time (RT) were calculated for the responses to each of the sounds, separately by category (speech, music, and environment and by stimulus position (1–8)). Hits were counted when responses occurred 100–1,100 ms from the onset of the stimulus. The mean HR was calculated as the total number of correctly identified stimuli divided by the number of stimuli in each category for each position. RT was calculated for each sound from sound onset. Means were derived for each stimulus category, in each position separately.

ERP Data: The filtered EEG was segmented into 4,500 ms epochs, starting from 200 ms pre-stimulus and ending 4,300 ms post-stimulus onset from position 1 for the switch category to display ERP responses consecutively in positions 1–4, and from the onset of position 5 for the repeat category to display ERP responses consecutively in positions 5–8. Due to the length of these epochs, ocular artifact reduction was performed on all participants using Neuroscan EDIT software. This Singular Value Decomposition transform method is used to identify the blink component. From the continuous EEG, a file was created that reflected the spatial distribution of the blink and then used to remove the blink. The blink-corrected data were then baseline-corrected across the whole epoch (the mean was subtracted at each point across the epoch). After baseline correction, artifact rejection criteria were set to ±75 mV. On average, 89% of all trials were included.

To measure mean amplitudes, the peak amplitude of each of the ERP components was visually identified in the grand-mean waveforms, in each condition separately, at the electrode with greatest expected signal-to-noise ratio for each component based on previous literature (Näätänen and Picton, 1987; Friedman et al., 2001; Fonken et al., 2020). Thus, we used the Pz electrode to measure the P3b component, the Cz electrode for the P3a component, and the Cz electrode for the N1 component. The peak latency in the grand-averaged waveforms were used to obtain mean amplitudes for statistical comparison. Mean amplitudes were calculated using a 50 ms interval centered on the grand-mean peak, for each ERP component, separately for each stimulus category and position, in each condition, for each individual participant.

For behavioral data (HR and RT), separate two-way repeated measures ANOVA with factors of category (speech/music/environment) and position (1–8) to determine effects of category switch and category repetition. For event-related potentials (N1/P3a/P3b), separate two-way repeated measures ANOVA with factors of category (speech/music/environment) and position (1–8) were calculated to determine effects of category switch and category repetition on the mean amplitude of the ERPs. In cases where data violated the assumption of sphericity, the Greenhouse–Geisser estimates of sphericity were used to correct the degrees of freedom. Corrected p values are reported. Tukey’s HSD for repeated measures was conducted on pairwise contrasts for post hoc analyses when the omnibus ANOVA was significant. Contrasts were reported as significantly different at p < 0.05. Effect sizes were computed and reported as partial eta squared (η²_p). Statistical analyses were performed using Statistica 13.3 software (Tibco).

Habituation is defined as a reduction in response to repeated stimulation that is not due to physiological effects, such as neural fatigue or adaptation (Magnussen and Kurtenbach, 1980; Rankin et al., 2009; Schmid et al., 2014). In the current study, we demonstrate a form of habituation that cannot be attributed to repetition suppression or neural fatigue. Conceptual categorical habituation was demonstrated by a dramatic decrease in the magnitude of the P3b amplitude after a single repetition of any of the categorical stimulus during active task categorization (Figures 2, 3). The delta in P3b amplitude from position 1 to position 2 was remarkable considering that the pattern of categorical stimuli occurred within an ongoing sequence of sounds, with no demarcation of when the grouping of category repetitions or switches were occurring. Further, the sounds themselves did not repeat, precluding stimulus-driven factors that could drive the response reduction by sensory adaptation or neural fatigue. The reduction in the magnitude of the neural signal after one repetition is notable because no two identical stimulus tokens were presented successively; the category repeated but not the physical stimuli. Accordingly, the reduction cannot be explained by stimulus-specific repetition suppression and reflects a neural habituation to the repetition of a conceptual category.

There was no categorical habituation when participants were watching a movie. It is well documented that when successive stimuli are acoustically unique, neural repetition suppression would not be expected (Näätänen and Picton, 1987; Budd et al., 1998; Grill-Spector et al., 2006). There was no conceptual “repetition suppression” when participants watched a movie. Thus, these results demonstrate that the conceptual categorical aspect of the stimuli was not automatically processed during passive listening. Habituation by repetition was only initiated when attention was focused on the sounds and a semantic categorical task was performed.

Expectation effects were observed only when the listener performed the categorization task with the sounds; not when they watched a movie. No explicit instructions were provided to participants about the stimulus structure, and the patterned structure was irrelevant to both tasks. However, implicit expectations could be formed by the regularity of the stimulus structure, in which the listener could expect a category switch after four successive categorical sounds most of the time. The larger P3b amplitude elicited by the categorical switch stimuli (position 1) demonstrates an implicit expectation that a category switch was likely to occur (a target switch). It should be noted that it was not possible to build up 100% expectation for a category switch because 30% of the time, rather than switching category, stimuli from the same category continued for a second successive group of four. Implicit category learning also influenced task performance. RT was slowest for position 1 stimuli: RT was 150 ms shorter to the second token of the within-category repetition. The faster responses time occurred for all the within-category stimulus repetitions (positions 2–8). The dramatic decrease in RT after only one repetition of a categorical stimulus is consistent with the substantial reduction in P3b amplitude after one categorical stimulus repetition (position 2) (Figure 3, compare positions 1 and 2). This is remarkable when considering that reaction time is faster to a repeated event than to a non-repeated event (Smith, 1968); that is, when it is the same physical stimulus. Here we show a reduction in RT to a conceptual repetition. The reduced response for position 2 stimuli indicates that the pattern of category repetition in the structure of the sequence was implicitly learned while performing the task. Implicit learning led to the knowledge that the category of stimuli would repeat after the switch position, even though that it was not the same physical sound token. The slower RT in position 1 and faster RT for positions 2–8 is consistent with modulation of the P3b amplitude, which was smaller after one category repetition and remained at the small amplitude until the next category switch. There was also an indication of implicit expectation in the longer RT at position 5, where a category switch may have been expected. However, this was not significantly reflected in the ERPs, likely due to the lower probability of a repeat than a switch.

In contrast, there was no evidence of implicit learning associated with the category switch when the listener watched a movie. The P3a amplitude did not differ as a function of position or category. The amplitude in position 1 was no different than that in any other position. Thus, a robust P3a was elicited by each successive stimulus token, with no indication by change in its magnitude that the brain detected a pattern of conceptual category repetitions. There was no categorical “repetition suppression.” Finding no index of implicit expectation, diverges from previous studies that have shown that stimulus repetition can build strong expectations and influence the brain response without attention focused on the sounds (Todorovic et al., 2011). However, our stimulus design is unique and may explain the differences in our results. The current study design differs from previous studies in two important ways. The repetition pattern of four sound tokens from the same category (speech, music, or environment) was comprised of four unique sound tokens from the category. For example, the listener may have heard the spoken words “peace” – “hello” – “yes” – “wonder” as the four-token repetition for one group in the speech category. All the sounds were different from each other. Therefore, identification of category repetition could not occur based on stimulus-driven features or acoustic characteristics of the sounds (Moskowitz et al., 2020). Secondly, expectations were not 100% predictable, that is, the category switch after the presentation of four sounds was not fully predictable; 30% of the time the category repeated. Consequently, during the passive condition, while attention was focused on reading captions and watching a movie, there could be some uncertainty about the regularity of the categorical aspect in the stimulus sequence, especially because the stimulus tokens themselves were not repeated, and attention was not actively monitoring the structure of the sound presentation. In addition, the structure of the sound sequence was irrelevant to performing the task. Thus, we conclude that attention focused onto the sounds with the intention to identify category membership was a key factor enabling expectations to be implicitly derived from the stimulus sequence.

A surprising result of the study was that a speech bias was observed only during active listening. Response times were faster and ERP amplitudes were larger to speech category tokens during task performance. There was no categorical effect when listeners were passively listening and watching a movie. The automatic involuntary orienting response (indexed by the P3a component) did not differentiate speech from the other categories at any position (Figure 2, Cz electrode, gray traces), whereas the P3b amplitude did differentiate speech (Figure 2, Pz electrode, black traces). Moreover, there was an attentional orienting response to the sounds in both the active auditory and the passive auditory conditions (Figure 2, Cz electrode, left panel, compare black and gray traces). However, with the active auditory task, there was an additional P3b component elicited consistent with target detection. There was no P3b elicited in the passive auditory when there was no auditory task. Thus, only with attention focused on a task with the sounds, was there evidence that the higher-level categorical aspects of signal differentiation. That is, differentiation of the speech signal from other music and environmental sounds was only evident when attention was used to categorize the sounds. This is notable because there is considerable evidence from infancy showing that speech is processed differently from other environmental sounds (Eimas et al., 1971; Pisoni, 1979; Murray et al., 2006; Vouloumanos and Werker, 2007; Agus et al., 2012; Gervain and Geffen, 2019). Recent evidence, however, has suggested that speech may only show an ‘advantage’ under specific listening or task situations (Moskowitz et al., 2020). In previous studies showing a speech bias, this issue of attention may have not come to light because stimulus categories were not separated by unique tokens. Certainly, one can detect speech passively and unattended speech can alert our attention (e.g., the sound of your name being called) (Navon et al., 1987). However, the current results indicate that when attention is not directed towards the sounds, the acoustic characteristics that distinguish speech from other environmental sounds are not automatically discriminated as a special category when there is a complex mixture of sound categories occurring. Our results indicate that attentional control is required to process the higher-level aspects of the speech signal, to extract the conceptual category (speech, music, or environment) when there are a variety of complex sounds. Speech may not be treated as a distinct or separate category without an active task and attention to the sounds. A question that remains is how specific the task must be to the conceptual process for it to alter the neural response; would performing a task not involving categorization also show no category effects?