|Home | About | Journals | Submit | Contact Us | Français|
In reverberant environments, the brain can suppress echoes so that auditory perception is dominated by the primary or leading sounds. Echo suppression comprises at least two distinct phenomena whose neural bases are unknown: spatial translocation of an echo toward the primary sound, and object capture to combine echo and primary sounds into a single event. In an electroencephalography (EEG) study, we presented subjects with primary-echo (leading-lagging) click pairs in virtual acoustic space, with inter-click delay at the individual's 50% suppression threshold. On each trial, subjects reported both click location (one or both hemi-fields) and the number of clicks they heard (1 or 2). Thus, the threshold stimulus led to two common percepts: Suppressed and Not Suppressed. On some trials, a subset of subjects reported an Intermediate percept, in which two clicks were perceived in the same hemi-field as the leading click, providing a dissociation between spatial translocation and object capture. We conducted time-frequency and event-related potential (ERP) analyses to examine the timecourse of the neural mechanisms mediating echo suppression. Enhanced gamma band phase synchronization (peaking at ~40 Hz) specific to successful echo suppression was evident from 20 – 60 ms post-stimulus. N1 latency provided a categorical neural marker of spatial translocation, while N1 amplitude still reflected the physical presence of a second (lagging) click. These results provide evidence that: 1) echo suppression begins early, at the latest when the acoustic signal first reaches cortex, and 2) the brain spatially translocates a perceived echo before the primary sound captures it.
Many everyday environments are highly reverberant, yet we often fail to hear echoes. This complex perceptual phenomenon (Wallach et al., 1949), known as the precedence effect or echo suppression, improves our ability to localize and identify sounds of interest (Freyman et al., 1999). It requires at least two mechanisms: spatial translocation and object capture (Blauert, 1999; Litovsky et al., 1999; Litovsky and Shinn-Cunningham, 2001). Spatial translocation, occurring over echo delays of ~1–9 ms for clicks, moves the perceived location of an echo toward the leading sound's location (Litovsky et al., 1999). Object capture, occurring at the shorter translocated delays (~ 1–5 ms for clicks), combines the leading and lagging sounds into a single, fused auditory object (Litovsky et al., 1999). Interestingly, at delays just above object capture threshold, an echo can be heard but at a location transposed toward the leading sound (Pecka et al., 2007). That is, spatial translocation can occur without object capture, providing a key dissociation between the mechanisms (Yang and Grantham, 1997). As echo delay increases further, both leading and lagging sounds are heard distinctly at their veridical locations.
Although the psychoacoustics associated with echo suppression have been well-described, its underlying neural mechanisms are poorly understood. The cochlear nucleus and inferior colliculus (IC) have been suggested by animal studies as the first levels where echo suppression begins (e.g. (Hafter et al., 1988; Wickesberg and Oertel, 1990; Yin, 1994; Fitzpatrick et al., 1995; Litovsky, 1998; Litovsky and Yin, 1998a, b; Pecka et al., 2007)), consistent with a human IC lesion study (Litovsky et al., 2002). In contrast, electroencephalography (EEG) studies in humans have shown that auditory brainstem responses index the physical presence of an echo regardless of perception (Damaschke et al., 2005), while late cortical potentials, including mismatch negativity (Damaschke et al., 2005) and object-related negativity (Sanders et al., 2008), may reflect perceptual consequences of echo suppression. Another recent study (Spierer et al., 2009) suggests that right temporo-parietal cortical activity is associated with a fused percept. Taken together, these electrophysiological studies leave a large time window between brainstem firing and late cortical potentials, wherein the precise timing and neural mechanisms underlying echo suppression remain unknown. Furthermore, the neural processes related to spatial translocation and object capture have never been dissociated.
In the present study, we used EEG to characterize the neural timecourse of echo suppression. Subjects listened to click pairs in virtual acoustic space, with inter-click delay calibrated to each subject's threshold, yielding three distinct perceptions of physically identical stimuli: Suppressed (translocation and object capture), Intermediate (translocation without object capture), and Not Suppressed (neither translocation nor capture). Two further conditions, Single click and obvious Double click, were perceptually or physically comparable to the threshold conditions, providing key reference points for the EEG analysis. We analyzed the EEG for changes in latency and amplitude of event-related potentials as well as spectrotemporal power and inter-trial phase-locking. We hypothesized that only physical stimulus properties would be manifest in the earliest EEG responses, followed by neural evidence of echo suppression mechanisms. Still later, we should observe the perceptual correlates of spatial translocation and, finally, object capture.
The protocol described here was approved by the Institutional Review Board at the University of California, Davis. All subjects gave written, informed consent before participating in the study and were paid for their participation.
All participants were right-handed, reported no neurological disorders, had not used any psychoactive medications within the month prior to their participation, and had normal pure-tone (200–12500 Hz) hearing thresholds. Of the 15 subjects completing the study, 13 (5 males, mean age = 22.3 years, SD = 2.8 years) had adequate behavior and EEG signal quality. Two subjects could not be included in the analysis due to too few trials after artifact rejection or poorly balanced behavior (trial numbers) among Suppressed, Intermediate, and Not Suppressed percepts.
Subjects were separated into two groups based on stimulus configuration, left or right leading hemi-field, as explained below. Five (3 males) of the 13 subjects completed the experiment for both stimulus configurations in order to test for intrasubject differences in echo threshold and electrophysiological results between hemi-fields. For these five subjects, EEG data for left- and right-leading sounds were collected in separate sessions on different days. Since the EEG analyses disregard overall mean differences across subjects (repeated measures analysis of variance, see below), each subject's two sessions were considered independently. Therefore, each configuration consisted of 9 subjects (4 males per group) for a total of 18 complete datasets.
Head-related transfer functions (HRTFs) were created for each subject in a sound-dampened chamber using AuSIM in-the-canal microphones (AuSIM, Inc., Mountain View, CA) and a Tannoy (Precision 6, Tannoy Ltd., UK) loudspeaker placed in the horizontal plane ~1.2 m away. Subjects were seated in a swivel chair, and white noise (3s duration) was recorded when subjects were positioned with the loudspeaker 45° to the left and 45° to the right of midline. HRTFs were estimated in the frequency domain, by dividing the magnitude and subtracting the phase of the fast Fourier transform (FFT) of the recorded sound in each ear relative to the FFT of the presented sound. From this HRTF, a head-related impulse response (HRIR) function was obtained by inverse FFT, and for practical purposes, the HRIR was truncated to the first 3000 time samples (i.e. filter order). This HRIR was then convolved with the stimuli.
All stimuli were biphasic clicks, consisting of 12 samples (6 positive and 6 negative) at a sampling rate of 96 kHz (0.125 ms long). The clicks were filtered with each subject's HRTF to place them in virtual acoustic space (see (Wightman and Kistler, 1989) for more details). Click pairs contained a leading click at either + or −45° followed by an equally intense lagging click in the opposite location. Subjects were randomly assigned to one of two groups (left-leading or right-leading) prior to the experiment. For the left-leading group, the leading sound was always presented from −45° (left) and the lagging sound from +45° (right) in virtual acoustic space (and vice versa for the right-leading group). All stimuli were presented at a comfortable listening level, with peak approximately 98 dBA SPL.
Trials consisted of a series of conditioning click pairs followed by a pause, a test click pair, and a cue for subjects to respond. The conditioning click train was used to prevent any one trial from affecting perception in subsequent trials, since short-term auditory context strongly affects the probability of echo suppression (Clifton, 1987; Clifton and Freyman, 1989; Clifton et al., 1994). Conditioning click trains consisted of 11 click pairs or single clicks (identical to the test pairs) and lasted ~1.8 seconds. On each trial, subjects listened to the conditioning click train, followed by a 1.6 second pause and a test click or click pair (identical to the one click or click pair in the conditioning train). There was about 0.8 second of silence between the test click and the conditioning train of the subsequent trial. Thus, the total length of each trial was 4.2 seconds.
Subjects were seated in a comfortable chair in a sound-dampened chamber, held the response keyboard on their lap, and faced a computer screen. Presentation software (NeuroBehavioral Systems, Inc., Albany, CA) was used to control stimulus timing and to present the auditory and visual (fixation cross) stimuli. Auditory stimuli were delivered through shielded ER-4 earphones (Etymotic Research, Inc, Elk Grove Village, IL). Participants visually fixated on a white cross in the center of the computer screen, and the cross flashed green at the start of the conditioning click train in each trial. Subjects were instructed to wait for the green flash before making any button responses for the trial they just heard. This prevented motor cortical activity from contributing to the EEG signals of interest. Upon seeing the green crosshair, subjects made two button responses in succession. First, they indicated with the right forefinger or middle finger whether they heard a click in one or both hemi-fields. They were told before the task that they would always hear a click on the left side (for left-leading subjects) or on the right side (for right-leading subjects). Second, they indicated if they heard either one or two clicks with a button press of their left forefinger or middle finger. They were further instructed that hearing a click on both sides always counted as two clicks, even if they sounded simultaneous. All subjects completed a practice session with feedback on each trial before beginning the behavioral calibration experiment.
In this session, subjects were presented with three conditions (as shown in Figure 1) in pseudorandom trial order: 1) Single clicks in the leading location, 2) Threshold click pairs (i.e. click pairs presented with a delay equal to the subject's echo threshold), and 3) Obvious double clicks (35 ms lag). Each subject completed a 20-minute calibration to determine his/her echo threshold, defined as the delay at which the subject reported the lagging sound on about 50% of the Threshold click trials. The calibration algorithm used an adaptive one-up-one-down procedure, based on the subject's responses (starting at 5 ms echo lag, consisting of 0.5 ms steps, and comprised of 290 trials). Each subject's echo threshold was determined as the most frequently occurring echo delay (on Threshold trials) during the second half of the calibration. This threshold from the behavioral session became the starting point for the EEG session abbreviated calibration.
The EEG session consisted of a 10-minute echo threshold recalibration, placing the cap and electrodes on the subject's scalp, and 8 15-minute blocks of experimental recording time. The calibration was done to confirm the echo threshold assessed during the behavioral session for each subject. After the calibration, the echo delay for Threshold trials was chosen for the duration of the EEG session. During the EEG recording, we verified that the subjects' suppression rate (or proportion of Suppressed to Not Suppressed trials) after each block was approximately 50:50. In some cases, a subject's suppression rate deviated substantially (e.g. 20:80) from this optimal ratio after the first block of EEG recording, so we adjusted the threshold to account for this, and continued recording. EEG analyses only used data collected with this new threshold.
Data was collected using high-density (128 channel) EEG (BioSemi system, BioSemi, Amsterdam) at a sampling rate of 1024 Hz. The EEG experiment lasted a minimum of two hours, and subjects were given breaks between blocks. Each block consisted of 215 trials, and the task was similar to the previous behavioral calibration tasks. However, instead of three stimulus conditions, there were five during the EEG experiment, also presented pseudorandomly: 1) Single clicks (in the leading click location), 2) Threshold click pairs, 3) Obvious double clicks (35 ms lag), 4) Single click catch trials (single click conditioning train, but no test click), and 5) Threshold pair catch trials (threshold click pair conditioning train, but no test pair). Subjects did not respond on catch trials, and the fixation cross did not flash green after these trials. Catch trials were used to ensure that no sustained brain activity from the conditioning train overlapped with the later activity locked to the test click. Other than the addition of catch trials, the task was the same as in the behavioral session.
EEG data preprocessing was done in MATLAB (The Mathworks, Inc., Natick, MA) using both EEGLAB software (Delorme and Makeig, 2004) and in-house code. First, each dataset was downsampled to 512 Hz, re-referenced to the average reference (i.e. the average activity of all channels was subtracted from each channel), high-pass filtered at 0.5 Hz (zero phase FIR), and separated into epochs (segments) spanning −1 sec to +0.5 sec, relative to the first test click on each trial. Thus, each epoch was 1.5 seconds long. There was one epoch for each trial of recorded EEG data, so each subject recorded a maximum of 1720 epochs (trials) prior to artifact rejection. Then, all epochs were baselined (normalized) to the average value measured during the 100 ms preceding the test click. All trials were then entered into Independent Components Analysis (ICA) (as implemented by EEGLAB (Makeig et al., 1997)), using 50 principal components. Canonical eye-blink and lateral eye movement independent components were conservatively removed from the data by visual inspection of each component map. After combining all non-artifactual ICA components, all trials were re-baselined (re-normalized) to the 100 ms preceding the test click. Artifact rejection removed epochs with data exceeding ±100 μV during the period −300 to +500 ms. Rejected epochs were removed from all 128 channels. Trials were then sorted according to stimulus type and the subject's responses. As shown in Figure 1, EEG trials were separated into seven perceptual categories: Single (single click), Suppressed (full suppression of the threshold echo), Intermediate (2 clicks in the leading hemi-field), Not Suppressed (perception of the echo), Double (obvious double click pair), Catch Single (single click catch trial), and Catch Threshold (threshold pair catch trial). This epoched data (i.e. data that has been segmented and separated by condition) then was used in both an Event-Related Potential (ERP) analysis and a time-frequency analysis, as detailed below.
The mean ERP waves of each perceptual condition were low-pass filtered at 20 Hz using a zero-phase finite impulse response (FIR) filter. These filtered, mean waves for each condition were averaged across all subjects to obtain the group mean waves for each condition.
We examined the amplitude and latency differences among conditions (Single, Suppressed, Not Suppressed, and Double) for three auditory ERP components (P1, N1, and P2). To analyze effects of laterality of the leading click in the P1 and N1, 10 left frontal channels and 10 right frontal channels (as shown in Figure 2B), at which the P1 and N1 reached their maximum and minimum values respectively, were averaged and the peak latencies within each condition were identified manually based on the group average (N = 18) of these 20 channels. These latencies were then used in an automated windowing algorithm, which recorded the local maximum (P1) or minimum (N1) amplitude and latency within 15 samples (~29.3 ms) on either side of the group peak latency for each subject within each condition and each side (ipsilateral or contralateral to the leading click location). Similarly, the peak latency for the P2 was based on the group average of Cz and its adjacent 5 channels (see Figure 2B). The local maximum amplitude and latency for each subject within each condition was automatically chosen. For ERP effects that revealed differences between Suppressed and Not Suppressed, we also analyzed the subset of subjects (N = 8) who reliably (≥ 50 trials post artifact rejection) reported the Intermediate percept.
Time-frequency analysis was performed using EEGLAB, which generates spectrograms for both inter-trial phase coherence (ITPC) and event-related spectral perturbation (ERSP). ITPC shows the extent to which neural oscillations are consistent in phase from one trial to another, within each channel. In other words, the term “phase-locking” as used in this article refers to the trial-to-trial consistency of the phase of a particular frequency band relative to the stimulus onset (first click). Thus, the phrase “enhanced phase-locking” indicates that the phase of a certain frequency within the EEG signal had a tighter temporal relationship with stimulus onset for one condition compared to another. ERSP, on the other hand, reflects a relative change in power over time following the stimulus, regardless of phase. Subjects (N = 2) who had fewer than 90 trials in any one condition were not included in the main time-frequency analyses, leaving 16 datasets. First, artifact rejection was repeated with the same threshold (+/−100 μV) but an earlier starting point of −1 second, because a longer baseline period is necessary for low-frequency ERSP baselining. A sliding Hanning-windowed 2-cycle sinusoidal wavelet (short time discrete Fourier transform (DFT)) of the EEG signal was used, in which the frequency increments were 2 Hz and the average step size was 6.23 ms. The sliding window size was 128 samples (250 ms, 2 cycles) at the lowest frequency (8 Hz), and it decreased linearly, while number of cycles increased with frequency, resulting in approximately 12 cycles at the highest frequency (100 Hz).
We conducted a time-frequency ROI analysis to determine if enhanced phase-locking during echo suppression was suppression-specific or reflected perception. Note that the time-frequency ROI is not a spatial ROI (as in some neuroimaging analyses), but rather an analysis of a specific time-frequency region within an EEG spectrogram. As mentioned above, we analyzed two types of spectrograms: (1) ITPC and (2) ERSP. Using the same ROI for ITPC and ERSP, we can examine whether differences between conditions in inter-trial phase-locking are also associated with differences in spectral power. ITPC generally returns Phase-Locking Factors (PLFs) at each time-frequency bin; PLF is the dependent measure in an ITPC spectrogram and ranges from 0 to 1, where 1 represents perfect phase-locking and 0 reflects phase de-synchronization (Tallon-Baudry et al., 1996). All ITPC values were converted from Phase-Locking Factors (PLFs) to Rayleigh's z value using the following equation: z = n*PLF2, where n equals the number of trials for that subject and condition (Fisher, 1993). (Note that Rayleigh's z value should not be confused with a z-score or normalized variate.) We conducted permutation tests (as detailed below) on the z value data to contrast phase-locking activity during Suppressed and Not Suppressed trials. From the thresholded (p < 0.05) ITPC contrast between Suppressed and Not Suppressed averaged across all channels, two main activation clusters, in which phase-locking during Suppressed was greater than during Not Suppressed (as depicted in Figure 5A), were chosen for an ITPC region of interest (ROI) analysis. The boundary for the early gamma ROI included 30 – 60 Hz from ~ 0 – 82 ms post-stimulus. The alpha/beta ROI included 8 – 28 Hz from ~ 50 – 212 ms.
The following steps describe the ITPC ROI analysis: (1) A box was drawn around each cluster to use as the time-frequency ROI mask for each channel. (2) Channels in which the ROI mask had 50 or more significant time-frequency bins were used. (3) Within these channels, each subject's z values at each of the significant group contrast pixels were returned. This step was done separately for each condition, but used the same ROI mask. (4) For each condition, each subject's z values were averaged. Steps 3 and 4 were repeated for the ERSP spectrograms to determine if changes in spectral power occurred at the same time-frequency bins as in the ITPC spectrograms. Note that the ERSP values were not converted into Rayleigh's z value since ERSP does not have a circular distribution.
Statistica software (version 8, StatSoft, Inc., Oklahoma) was used to compute statistics, particularly t-tests for behavioral results and repeated measures analysis of variance (RMANOVA) and post-hoc Fisher's Least Significant Difference (LSD) tests for all EEG results. The proportion of Suppressed and Not Suppressed trials remaining after artifact rejection of the EEG data was used to calculate suppression rate on Threshold trials. Independent t-tests were used to compare the left- and right-leading groups' suppression rates and echo thresholds. For the ERP analysis, a 4 (Condition) × 2 (Side) × 2 (Group) RMANOVA was done for the latency and amplitude of the P1 and N1. For the P2 amplitude and latency, a 4 (Condition) × 2 (Group) test was done. Since P2 topography was centralized, unlike P1 and N1, laterality effects were not examined.
For the time-frequency analyses, permutation tests were used for statistical thresholding of the main contrast (Suppressed – Not Suppressed) at the group level, as described in detail by (Chau et al., 2004) and (Shahin et al., 2008). Briefly, permutation methods are nonparametric, in that they determine the null distribution by resampling the data rather than assuming an analytic form for the population distribution. This null distribution is derived by randomly relabeling a subject's data with a condition, such as Suppressed or Not Suppressed, and quantifying the ITPC or ERSP differences between these conditions. Each random assignment of data therefore produces a chance (null) distribution of ERSP or ITPC differences across electrodes and time-frequency bins, based on the pre-stimulus period. This distribution reflects the null hypothesis that oscillatory activity does not differ between conditions. To handle the problem of multiple comparisons, permutation tests were applied based on the null distributions of the maximum values obtained in repeated resamplings of the data (Holmes et al., 1996). This maximal null distribution was used to determine the threshold for a given p value (here, p<0.05).
RMANOVAs were also done to compare phase-locking and spectral power within each ROI. The Catch trials acted as a control, to measure any activity from the conditioning train that sustained through the test click. For each time-frequency ROI, the Rayleigh's z values (for ITPC) or spectral power values (dB, for ERSP) were averaged for the Catch Threshold and Catch Single conditions; this mean was subtracted from the z values or spectral power of the other four conditions (Single, Suppressed, Not Suppressed, and Double). For these four conditions (Single – Catch, Suppressed – Catch, Not Suppressed – Catch, and Double – Catch), 4 (Condition) × 2 (Group) RMANOVAs were computed for ITPC z values and spectral power. Since the creation of the ROI clusters were based on differences between phase-locking of Suppressed and Not Suppressed trials, the ITPC RMANOVA is likely significant. However, we were mainly interested in the post-hoc LSD tests to compare phase-locking among all four conditions, not only differences between Suppressed and Not Suppressed. On the other hand, because the time-frequency bins in the spectral power plots examined were not based on pre-existing differences in spectral power, neither the RMANOVA nor post-hoc LSD tests for the ERSP analysis was biased. In both ITPC and ERSP, the post-hoc LSD tests showed whether an effect was specific to echo suppression or generally related to perception, as discussed below.
Since only 8 subjects reliably reported the Intermediate percept on Threshold trials, these trials were not included in the overall suppression rate. Thus, each subject's overall suppression rate is their proportion of Suppressed to Not Suppressed responses. There was no significant difference (p = 0.93) in overall suppression rate between left- (Mean = 59.0%, SD = 13.5%) and right-leading (M = 59.6%, SD = 16.1%) groups. Thus, these proportions give a balanced sample of both Suppressed and Not Suppressed trials for the EEG analyses. Also, the echo thresholds for the two groups were very similar (p = 0.83). The left-leading group's mean echo threshold was 6.33 ms (SD = 2.02), while the right-leading group's mean was 6.11 ms (SD = 2.33).
A standard ERP analysis was done to examine interactions among either amplitude or latency of auditory evoked potentials (P1, N1, and P2), condition (Single, Suppressed, Not Suppressed, or Double), and laterality of activity relative to the leading click hemi-field (for P1 and N1). However, no main effects or interactions were found involving laterality of activity for either component, so all results reported collapse activity across hemispheres (see Figure 2B). We therefore used repeated measures analyses of variance (RMANOVAs) to test whether each component's latency or amplitude, averaged within subject (see Methods), showed a main effect of condition.
We expected a series of neural responses which index different aspects of echo suppression. Four possible relationships among conditions, or profiles, and their corresponding interpretations are presented in Figure 3: (1) Since the Single condition was the only condition without a lagging “echo” click, difference in activity between the Single and the other conditions indexes the physical presence of the “echo” regardless of perception (Figure 3A). (2) If Suppressed activity differs from the other conditions, it reflects the echo suppression mechanism (Figure 3B). Finally, if both Single and Suppressed activities are similar, both Not Suppressed and Double are similar, and these two pairs are distinct, this pattern reflects the perceptual consequences of echo suppression. However, only with the Intermediate percept can we attribute this effect unambiguously to either spatial translocation or object capture. Namely, (3) if the Intermediate value is similar to Single and Suppressed values, the effect indexes spatial translocation (Figure 3C). (4) However, if the Intermediate, Not Suppressed, and Double values are similar, the effect reflects object capture (Figure 3D).
We analyzed the P1 amplitude and latency (see Figure 4A) to determine if there were any early effects in the ERP related to echo suppression. P1 amplitude showed a main effect of condition (F(3,48) = 3.30, p = 0.028), such that the P1 of the Double click was smaller than the other three conditions (post-hoc, p < 0.05). Since the lead and lag were separated by 35 ms in this condition, it is probable that the P1 responses to each click destructively interfered with one another, causing a summed component to flatten and broaden. However, the lead and lag of the threshold pair were only separated on average by approximately 6 ms; thus, the destructive interference would be minimal, resulting in P1 amplitudes similar to that of the single click. There was no main effect of condition on P1 latency (F(3,48) = 1.75, p = 0.17).
Whereas the P1 did not reveal any early perceptual or echo suppression-specific effects, perception was reflected in the N1. Both N1 amplitude and latency showed a main effect of condition (F(3,48) = 4.35, p = 0.0087; F(3,48) = 7.05, p < 0.001, respectively), as shown in Figure 4B. Post-hoc LSD tests showed that the amplitude for the Single click condition (−1.2 +/− 0.23 μV) was smaller than all other conditions (Suppressed (−1.4 +/− 0.20 μV), Not Suppressed (−1.5 +/− 0.23 μV), and Double click (−1.4 +/− 0.23 μV)). For the N1 latency, according to post-hoc LSD tests, there was no difference between that of the Single (128 +/− 5.9 ms) and Suppressed (126 +/− 4.4 ms) conditions or the Not Suppressed (133 +/− 4.9 ms) and Double (136 +/− 4.7 ms) conditions. Furthermore, the N1 occurred earlier for the Single and Suppressed trials than for Not Suppressed and Double trials. Thus, N1 amplitude reflects stimulus properties (same effect profile as in Figure 3A) and N1 latency reflects echo perception (one or two clicks heard, on one or both sides).
There was no main effect of condition on P2 amplitude (F(3,48)=1.01, p = 0.40). A main effect of condition (F(3,48) = 14.34, p < 0.001) on P2 latency was found, such that the P2 latency was smallest for Single (166 +/− 4.1 ms) and Suppressed (171 +/− 3.8 ms) trials and largest for Double (190 +/− 4.5 ms) click trials (see Figure 4C). Specifically, post-hoc LSD tests revealed a significant difference between the P2 latency of Double clicks and of the other three conditions. Also, P2 latency was significantly reduced for Single compared to Not Suppressed (175 +/− 3.9 ms) trials. The Suppressed trial latency was similar to that of both the Single condition (p = 0.23) and the Not Suppressed P2 latency (p = 0.24).
In order to distinguish between the mechanisms of spatial translocation and object capture (as illustrated in Figure 3C and D, respectively), we analyzed the effects that generally reflected perception (N1 and P2 latency) within the 8 Intermediate percept subjects. (See Figure 4D and E, respectively.) N1 latency showed a main effect of condition (F(4,24) = 4.38, p = 0.0084). Post-hoc LSD tests revealed that the N1 latencies of Suppressed (125 +/− 7.6 ms) and the Intermediate percept (128 +/− 5.3 ms) were both faster than those of Not Suppressed (141 +/− 7.3ms) (p < 0.01). Since subjects reported hearing sounds in only the leading hemi-field for Suppressed and Intermediate percept trials versus both hemi-fields for Not Suppressed trials, the decrease in N1 latency is related to spatial translocation rather than object capture. Also, for P2 latency, there was a main effect of condition (F(4,24) = 3.08, p = 0.035). P2 latencies for Suppressed (163 +/− 5.6 ms) and Not Suppressed (172 +/− 4.6 ms) were significantly different, but the P2 latency of the Intermediate percept (166 +/− 6.5 ms) was not different from either Suppressed or Not Suppressed, providing no clear distinction between translocation and capture mechanisms.
Since echo suppression is a rapid, time-sensitive phenomenon, a time-frequency approach may complement the traditional ERP analysis, for at least two reasons. First, neural processing can manifest as oscillatory activity, which is naturally suited to frequency-domain representation. Second, ERPs tend to measure effects only evident in trial-averaged signals, whereas time-frequency approaches first quantify effects in single trials, and then take the average of those effects. If neural timing and amplitude are ever uncoupled, condition differences in timing (phase) and amplitude (power) will be confounded in ERP averaging, but will remain distinct in time-frequency analyses. For instance, inter-trial phase coherence (ITPC) shows the degree to which neural activity is phase-locked to a particular event, such as sound onset, thus focusing on timing rather than power. We compared ITPC between Suppressed and Not Suppressed conditions, revealing two significant spectrotemporal clusters that were consistent across channels: an early gamma cluster (30 – 60 Hz from ~ 0 – 82 ms post-stimulus) and a later alpha/beta cluster (8 – 28 Hz from ~ 50 – 212 ms). These clusters were then used as time-frequency regions of interest (see Methods) to determine if enhanced phase-locking was specific to an echo suppression mechanism or reflected perceptual consequences, following the same logic discussed above and shown in Figure 3. To reiterate, an effect (here, enhanced phase-locking) that distinguishes Suppressed from the other conditions reflects the echo suppression mechanism (Figure 3B). On the other hand, if conditions are grouped by what subjects heard (one click for Single and Suppressed versus both clicks for Not Suppressed and Double), the activity indexes the perceptual consequences suppressing an echo (Figure 3C and D). We used RMANOVAs (N=16) to determine significant differences in phase-locking among conditions. The same ROI masks were used to examine spectral power within each condition. Results from this analysis are shown in Figure 5B and C.
In the early gamma phase-locking ROI, there was a significant main effect of condition (F(3,42) = 10.97, p < 0.0001). Post-hoc LSD tests showed that Suppressed had higher phase-locking than the other conditions (p < 0.001). There was also a main effect of condition within the low frequency phase-locking ROI (F(3,42) = 24.77, p < 0.0001). Similar to the first ROI, phase-locking during Suppressed trials was greater than all other conditions (p < 0.0001). In contrast to phase-locking, there was no main effect of condition for spectral power within the gamma ROI (F(3,42) = 1.43, p = 0.25), and only a strong trend in the alpha/beta ROI (F(3,42) = 2.86, p = 0.048), driven by decreased power for the Double relative to the Suppressed and Not Suppressed conditions (p < 0.05). Since phase-locking in both the early gamma and low frequency ROIs was enhanced only for Suppressed, the one condition when echo suppression was both required and successful, then this enhanced phase-locking likely indexes the mechanism underlying suppression (as represented in Figure 3B). Furthermore, the enhanced phase-locking was not accompanied by differential spectral power. Therefore, the echo suppression mechanism is reflected by increased temporal precision in neural firing both early (20 – 60 ms, ~40Hz) and for hundreds of ms following a stimulus (~50 – 212 ms, 8 – 28 Hz).
Since there were only 8 Intermediate percept subjects, some of whom had fewer than 90 trials in one or more conditions, there was not sufficient power for an independent time-frequency analysis. However, we conducted a supplementary analysis to determine if the echo suppression mechanism, evident in ITPC, reflected translocation alone or both translocation and object capture. The time-frequency ROI masks generated from the whole group analysis Suppressed – Not Suppressed contrast were used in this analysis. Despite low trial numbers of some subjects, the results were very similar to the whole-group results. There was a main effect of condition for both the early gamma and low-frequency ROIs (F(4,24) = 4.66, p < 0.01; F(4,24) = 6.10, p < 0.005, respectively). In both time-frequency ROIs, phase-locking was enhanced for only the Suppressed condition while the Intermediate condition was similar to Single, Not Suppressed, and Double (post-hoc LSD, p < 0.01; Figure 5 D and E). Thus, unlike the N1 latency effect, which reflected the perceptual consequences of echo suppression (since Single, Suppressed, and Intermediate were distinct from Not Suppressed and Double), this phase-locking enhancement is specific to the echo suppression mechanism. These results suggest that enhanced phase-locking occurs only during full echo suppression — that is, when the echo is both translocated and captured by the leading sound.
In this study, we characterize the temporal dynamics of echo suppression in humans using EEG. Using a subject-specific threshold echo delay, we compared activity related to three distinct perceptions (Suppressed, Intermediate, and Not Suppressed) that arose from the same physical stimulus. Single and obvious double clicks provided a controlled measure by which we distinguished EEG responses based on stimulus attributes, suppression mechanism, and perceptual consequences. The Intermediate percept was particularly informative in that it dissociated two key percepts, spatial translocation and object capture. We used both event-related potential (ERP) and time-frequency analysis methods to determine the following neural timeline for echo suppression: (1) the echo suppression mechanism is evident in enhanced early gamma followed by low frequency phase-locking, (2) perceptual spatial suppression is reflected in shorter N1 latency, while N1 amplitude still reflects the physical rather than perceptual presence of an echo, and (3) P2 latency may index the probability of fusing the echo and leading sounds or the interaction between physical stimulus properties and perception.
Evidence for the echo suppression mechanism was apparent very early, peaking from 20 to 60 ms post-stimulus. We observed enhanced low gamma (~40 Hz) phase-locking specific to successful suppression, peaking at the vertex and fronto-central channels. Based on the timing and topography of this enhancement, this activity is likely related to the auditory middle latency response (AMLR). The AMLR occurs after the auditory brainstem response (ABR) and consists of three main peaks: Na, Pa, and Nb (Goldstein and Rodman, 1967). The latency of the early phase enhancement matches that of the Na (~20 – 24 ms), Pa (~31 – 35 ms) and Nb (~42 – 50 ms) (Goldstein and Rodman, 1967; Neves et al., 2007). Since the P1 peaked at about 60 ms in this study, the later portions of this enhanced synchrony could partially reflect the P1 as well. However, the Na, Pa, and Nb usually peak at the vertex, matching the topography of this enhancement, whereas the P1 was strongest in frontal channels. Previous findings suggest that the Pa and Nb are generated in Heschl's gyrus and thalamo-cortical circuits (Picton et al., 1974; Hall III, 1992; Liegeois-Chauvel et al., 1994; Yvert et al., 2001). Furthermore, such evoked 40 Hz gamma band responses have been shown to be modulated by top-down or contextual processes, especially auditory expectations (Widmann et al., 2007; Schadow et al., 2009) of the sort that may build up during the conditioning click train. Thus, the echo suppression mechanism likely reflects recent acoustic context (built up during the conditioning click train) and is manifest, at the latest, in either thalamus or Heschl's gyrus.
This echo suppression mechanism, evidenced by enhanced inter-trial phase synchrony, continued in the alpha/beta frequency range from about 50 to 200 ms post-stimulus. Although the amount of low frequency inter-trial phase synchronization between trials may be proportional to the amplitude of ERP components, this enhanced phase-locking was not evident in the N1 or P2 amplitudes. This suggests that the time-frequency analyses used in this study are more sensitive to fine fluctuations in phase. Furthermore, enhanced phase-locking during echo suppression indicates that a higher degree of temporal precision in neural firing, and thus more aligned phase, is likely important for the mediation of full echo suppression, including both spatial translocation and object capture mechanisms. Under this interpretation, if the neural response mediating echo suppression does not reach this level of temporal alignment to the stimulus, suppression will likely fail. The enhanced low frequency phase-locking suggests that the echo suppression mechanism continues through the timeframe of the N1 and P2, during which the resulting neural correlates of perception emerge.
N1 amplitude and latency index stimulus attributes and perception, respectively. The single click contained half the acoustic energy compared to the other conditions, resulting in smaller N1 amplitude. The N1 amplitude thus faithfully encodes the acoustic presence rather than perceptual suppression of the echo. In contrast to amplitude, N1 latency reflects perceptual consequences of echo suppression. Specifically Single, Suppressed, and Intermediate conditions had faster N1 latencies than Not Suppressed and Double trials. Recall that when subjects reported the Intermediate percept, they suppressed most of the echo's directional information (translocation) since they heard two objects in the leading click's hemi-field while preserving the echo as a separate object (no object capture). Thus, the Intermediate percept provides a clear dissociation between spatial translocation and object capture. In contrast, during Suppressed trials, both translocation and object capture occurred, resulting in only one perceived object. But when subjects detected the echo (Not Suppressed), both aspects of suppression failed. In light of this, shorter N1 latency is a categorical marker of spatial translocation. The N1 in the present study is likely an overlap of neural activity from the N1b, which peaks at the vertex at ~100 ms (Vaughan and Ritter, 1970; Naatanen and Picton, 1987), and the N1c, which reaches its maximum amplitude in temporal electrodes from 138 to 155 ms (Perrault and Picton, 1984; Shahin et al., 2003; Bosnyak et al., 2004). An overlap of the N1b and N1c components explains both the late N1 latency reported here as well as the topography, as shown in Figure 2A.
Whereas N1 latency categorically represents spatial translocation, the P2 latency may reflect the additional consequences of object capture. P2 latencies increased with echo lag as follows: Single ~ Suppressed ≤ Intermediate ≤ Not Suppressed < Double. Since in general, the probability of object capture decreases with increasing echo lag, the P2 latency could index the ease or likelihood of capture. However, this view is complicated by the fact that Intermediate latency is between Suppressed and Not Suppressed rather than grouped with Not Suppressed. Alternately, the P2 latency may reflect an interaction among stimulus properties, translocation (which was already evident in the N1 latency), and object capture. This integration of perceptual outcomes could result in the graded P2 latency for Suppressed, Intermediate, and Not Suppressed. Recall, the echo is still clearly encoded as a second object in the N1 amplitude, suggesting that capture has not yet occurred by the time spatial translocation is evident. Furthermore, a recent EEG study (Sanders et al., 2008) showed that echo detection, and thus hearing two sounds, was indexed by Object-Related Negativity (ORN). The ORN is a negativity associated with perceiving two versus one sound that overlaps with the N1 and P2 (Alain et al., 2001). Taken together, the neural mechanisms underlying object capture likely occur after the N1, and thus after spatial translocation.
Our results therefore suggest that spatial translocation temporally precedes and may be necessary for object capture. Furthermore, the processes underlying spatial translocation and object capture may be mediated in communicating yet parallel pathways. This mechanistic dissociation recalls the large-scale organization of the auditory cortical system which, like the visual system (Ungerleider and Haxby, 1994), has dorsal and ventral pathways for processing spatial and non-spatial information, respectively (Rauschecker, 1998). Consistent with our results, two recent studies suggest that spatial processing in the dorsal auditory pathway occurs faster than acoustic feature analysis in the ventral pathway (Ahveninen et al., 2006; Altmann et al., 2007). Ahveninen et al. (2006) demonstrated that the “where” pathway was activated 30 ms before the “what” pathway, while Altmann et al. (2007) showed that changes in sound location were processed about 100 ms faster than changes in acoustic patterns. Since the brain must know if a sound is an echo before suppressing its directional information, basic acoustic features (e.g. spectral profile) must be evaluated to accurately label an echo. This rudimentary feature information may be passed from the ventral to dorsal stream so that the brain can spatially translocate the correct echo. Alternately, potential echoes could already be “tagged” in subcortical structures such as the inferior colliculus without ventral cortical stream involvement (Pecka et al., 2007). A third possibility is that the dorsal stream might rely on its own lower-fidelity feature representations to identify the echoes. In any case, the spatial translocation system must influence the final object representation, as we cannot perceive a fused auditory object originating from two distinct spatial locations. Thus, if spatial translocation is successful, the processes underlying object capture continue and usually (but not always) fuse the leading and lagging sounds into one object.
Finally, our design, with a build-up click train on every trial, differs from numerous studies that use only one click pair on each trial (see (Litovsky et al., 1999) for a review). The build-up of echo suppression strongly depends on short-term auditory context, which informs the brain's “model” or expectations about its acoustical environment (Clifton, 1987; Clifton and Freyman, 1989; Clifton et al., 1994). Thus, measuring electrophysiological markers of echo suppression following a single click pair in isolation would likely reflect not only echo suppression per se, but also the build-up process during which perception is not yet stabilized. In the present study, we were interested in echo perception once it had stabilized — that is, once the brain's expectations had been established. Since these two designs are aimed at different neural processes related to echo suppression, they will likely yield different yet complementary electrophysiological results.
The present study sheds light on the neural time-course underlying echo suppression. The first neural marker, enhanced gamma inter-trial phase synchronization, begins at about 20 ms post-stimulus and reflects the neural mechanism that predicts successful echo suppression. This suppression-specific phase coherence enhancement continues in low frequencies from approximately 50 – 200 ms, as correlates of spatial suppression and object capture emerge. By about 130 ms, suppression of the echo's spatial information is categorically indexed by the N1 latency. However, the acoustic attributes of the stimuli, specifically the presence of a second “echo” click, is still encoded by the N1 amplitude. Finally, P2 latency may reflect the ease of object capture or an interaction between stimulus properties and perception. Thus, cortical mechanisms underlying echo suppression are evident first, followed by correlates of spatial translocation and finally, object capture.
This research was supported by the National Institutes of Health/National Institute on Deafness and other Communication Disorders (LMM, grant number R01-DC08171). The authors wish to thank Jess Kerlin for insightful comments.