Auditory attention reduced ear-canal noise in humans, but not through medial olivocochlear efferent inhibition: Implications for measuring otoacoustic emissions during behavioral task performance

Otoacoustic emissions (OAEs) are often measured to non-invasively determine activation of medial olivocochlear (MOC) efferents in humans. Usually these experiments assume that ear-canal noise remains constant. However, changes in ear-canal noise have been reported in some behavioral experiments. We studied the variability of ear-canal noise in eight subjects who performed a two-interval-forced-choice (2IFC) sound-level-discrimination task on monaural tone pips in masking noise. Ear-canal noise was recorded directly from the unstimulated ear opposite the task ear. Recordings were also done with similar sounds presented, but no task done. In task trials, ear-canal noise was reduced at the time the subject did the discrimination, relative to the noise level earlier in the trial. In two subjects, there was a decrease in ear-canal noise, primarily at 1-2 kHz, with a time course similar to that expected from inhibition by MOC activity elicited by the task-ear masker noise. These were the only subjects with spontaneous OAEs (SOAEs). We hypothesize that the SOAEs were inhibited by MOC activity elicited by the task-ear masker. Based on the standard rationale in OAE experiments that large bursts of noise are artifacts due to subject movement, noise bursts above a sound-level criterion were removed. As the criterion was lowered and more high-and moderate-level noise bursts were removed, the reduction in noise level from the beginning of the trial to the time of the 2IFC discrimination became less. This pattern is opposite that expected from MOC inhibition (which is greater on lower-level sounds), but can be explained by the hypothesis that subjects move less and create fewer bursts of noise when they concentrate on doing the task. In contrast, for the six subjects with no SOAEs, in no-task trials the noise level was little changed throughout the trial. Our results show that measurements of MOC effects on OAEs must measure and account for changes in ear-canal noise, especially in behavioral experiments. The results also provide a novel way of showing the time course of the buildup of attention in ear-canal noise during a 2IFC task.


Introduction 47
Medial olivocochlear (MOC) efferent activity has long been hypothesized to facilitate hearing in 48 noise (Nieder and Nieder, 1970;Michel and Collet, 1993;. Many papers have 49 attempted to determine how MOC efferent activity affects hearing by measuring changes in 50 otoacoustic emissions (OAEs) as subjects performed an auditory task that was expected to elicit 51 efferent activity (e.g. Puel et al., 1988;Meric and Collet, 1994;de Boer and Thornton, 2007;52 Harkrider and Bowers). MOC activity reduces the gain of cochlear amplification and thereby reduces 53 OAEs, so OAE reductions provide information about efferent activation and its effects in the cochlea. 54 A key assumption in measuring OAEs during behavioral task performance has been that there is no 55 change in the background level of the random noise in the ear canal, so that any measured changes in 56 OAEs can be attributed to changes produced by MOC efferents. 57 In contrast to the assumption that ear-canal noise is not changed during a behavioral task, several 58 studies have reported such changes (de Boer, and Thornton, 2007;Walsh et al., 2014a;2014b, 2015. 59 Walsh et al. (2014;2015) reported that ear-canal random noise was reduced by selective attention 60 activating MOC efferents. In the Walsh et al. experiments, ear-canal noise was indirectly measured 61 during a 30 ms silent period by a double-evoked technique that yielded a measure termed a 62 "nonlinear stimulus frequency otoacoustic emission" or "nSFOAE" (Walsh et al., 2010). During both 63 auditory and visual tasks there was a reduction in ear-canal noise (i.e. a reduction in the nSFOAE) 64 relative to when the subject was presented the same stimuli but did not do a task (Walsh et al., 2014a;65 2014b, 2015). For an auditory task, the reduction was similar in both the attended ear and the 66 opposite ear. Walsh et al. hypothesized that cochlear-amplified random vibrations within the cochlea 67 created backward traveling waves that produced acoustic noise in the ear canal, and activation of 68 MOC efferents reduced cochlear amplification and therefore reduced the random noise within the ear 69 canal. 70 We have done experiments that allow us to measure changes in ear-canal acoustic noise during a 71 behavioral task. Our subjects did a two-interval-forced-choice (2IFC) level discrimination task on 72 monaural tone bursts in noise. During these tests we measured changes in click-evoked otoacoustic 73 emissions (CEOAEs) in the task ear, with the goal of assessing changes in MOC activation during 74 the behavioral task. Most relevant here is that we also measured the sound pressure in the ear where 75 no sound was presented, opposite to the task ear. These opposite-ear recordings provide an 76 opportunity to directly determine whether there was a reduction in ear-canal background noise sound 77 pressure during the behavioral task, and to measure its time course relative to the time when sounds 78 were presented and the subject made the 2IFC judgment. 79 Methods 80

Subjects 81
Eight subjects (aged 18-21 years; 2 male) participated in the experiments reported here. All subjects 82 had normal pure-tone audiograms (<15 dB HL at octave frequencies 0.5 to 8 kHz). Sounds were 83 presented and recorded using Etymotic Research ER10c acoustic assemblies, sampled at 25 kHz. The 84 acoustic outputs were monitored and calibrated frequently throughout the experiments. This study 85 was performed in accordance with MEEI, MIT and NIH guidelines for human studies. Informed 86 consent was obtained from all subjects. 87

Experimental methods 88
The experiments were designed to detect changes in CEOAE amplitudes brought about by efferent 89 activity, i.e. changes in CEOAEs from the beginning of a 2IFC trial to just after the stimuli to be 90 discriminated in the trial (the masking noise made it too difficult to measure CEOAEs during the 91 noise). We did both "active" runs in which the subject did the 2IFC task, and "passive" runs in which 92 the subject heard the same sounds but made no judgment. Since learning to do the 2IFC task might  93  cause a subject to continue to attend to the task sounds during passive trials, passive trials were done  94 first, before the subjects were told about their future task. Passive and active conditions were 95 typically done in separate sessions, where a "session" is defined as the time that a subject 96 continuously had the acoustic-assembly foam plugs in their ear canals. Removing and replacing the 97 acoustic assembly was considered a new session, whether it was a few minutes later or days later. 98 Since acoustic parameters such as the depth of insertion might change across sessions, direct 99 comparisons of the amplitudes of the ear-canal acoustic noise in active versus passive listening were 100 not done because such comparisons may not be accurate. However, the stimuli and their timing were 101 the same in passive and active trials so we can compare the time courses of ear-canal sound in 102 passive and active trials. 103 Sound stimuli were presented only in the task ear, which was the ear that had the most robust 104 CEOAEs in our initial tests. The subject's task was to detect which of two short tone bursts was 105 larger in amplitude. Both tone bursts were embedded in 50 dB SPL broad-band noise. only clicks (see Fig. 1). This was followed by 3 epochs that had the clicks plus 50 dB SPL, broad-117 band (0.1-10 kHz) frozen noise (the same in each epoch). The last two epochs with clicks and noise 118 also had a tone burst (15 ms plateau, 5 ms raised-cosine rise and fall times) that ended 45 ms before 119 the end of the epoch. After the tone-in-noise epochs there was an additional 400 ms epoch in which 120 there were only repeated clicks (the same as in initial epochs 1-10) (Fig. 1). Overall, the number of 121 400 ms epochs in each trial varied from 5 to 14, depending on the number of initial epochs. At the 122 end of each trial the subject indicated whether the first or second tone burst was higher in level by 123 pushing one of two buttons on a device on which their hand rested (usually this done was during the 124 last 400 ms epoch). To push the proper button, a subject only had to move one finger and did not 125 have to move their arm. We did not have subjects type on a keypad or touch a screen so as to 126 minimize subject motion. The next trial in the batch of 25 trials began 1 second after the button press 127 or end of the last epoch, whichever came later. 128 Spontaneous otoacoustic emissions (SOAEs) were measured once on each subject by recording the 129 ear canal sound in both ears simultaneously with no stimulus presented and the subjects instructed to 130 sit very still for this short measurement. On each ear, eight data buffers were obtained, each sampled 131 every 40 µs and 2.62 seconds long. Each buffer was individually fast-Fourier transformed and the 132 resulting amplitudes (phases set to zero) were averaged. Two subjects (323, 326) had SOAEs, as 133 judged by their having spectral lines that were >10 dB above the smoothed SOAE spectra. 134

Data analysis 135
Throughout each trial, sound was recorded continuously in both ears and stored for later processing. 136 The data for the present paper are from the ear opposite the task ear, except that the test for middle-137 ear-muscle (MEM) activation used the amplitudes of the clicks in the task ear. Before processing, the 138 opposite-ear data were filtered from 0.5-5 kHz by a zero-phase-change FIR digital filter. The 139 opposite-ear recordings were divided into 25 ms time spans-hereafter referred to as "spans"-140 corresponding to the times demarcated by the clicks in Figure 1. We measured the root-mean-square 141 (RMS) value of the sound in every time span. We visualized the amplitude distribution of the RMS's 142 from the spans in a batch of 25 trials-hereafter referred to as a "batch"-by binning the RMS values 143 into 300-bin histograms with the 100 th bin equal to the median value of the RMS distribution and bin 144 widths of 1% of the median value (Fig. 2). RMS values greater than three times the median value 145 were used later, but were omitted from the histogram. For most sessions, these RMS histograms had 146 narrow peaks and tails with higher RMS values (e.g. Fig. 2). For subsequent data analysis, a given 147 span was not used if its RMS value was above a rejection criterion RMS value that was a parameter 148 varied in our study. To find a criterion value, we first smoothed the histogram and then determined 149 an "upper-edge RMS" value, where the histogram fell to 50% of the peak. The difference between 150 the upper-edge-RMS and the peak RMS is termed the "Edge Width". The Edge Width, multiplied by 151 a user-chosen constant (the "Edge Multiplier), and added to the peak RMS value, defined the 152 rejection criterion RMS value. 153 The opposite-ear sound recordings were contaminated to varying degrees by crosstalk from the task-154 ear masker noise. This crosstalk was assessed from the difference between two ways of combining 155 pairs of span waveforms from different trials of a batch: (1) reversing the polarity of one waveform 156 of the pair and then averaging, or (2) averaging the waveforms without reversing either one (Fig. 3). 157 Since the frozen-noise masker was the same on every trial, reversing the polarity of one waveform 158 before averaging cancels the crosstalk contribution in the average. In contrast, if the ear-canal sound 159 is random noise, reversing the sign of a waveform before averaging makes no difference. The 160 difference in these two measures (each averaged over the time when the masker noise was on: epochs 161 11-13) and converted to dB SPL, yielded crosstalk levels averaging -22 dB SPL (range -31 to -10 dB 162 SPL). We compensated for the square-root-of-2 adjustment appropriate for averaging noise but not 163 appropriate for averaging the crosstalk signal. The task-ear masker noise was 50 dB SPL so the 164 crosstalk attenuation averaged 72 dB. In a few sessions, the crosstalk and/or other aspects of the 165 recordings were highly abnormal (differed by more than a factor of two from the other values on that 166 subject -perhaps the acoustic assemblies were not properly seated); these data were excluded from 167 our analysis. 168 To avoid masker-noise crosstalk from affecting the noise rejections, we used a two-step procedure to 169 exclude noisy spans. The procedure described below was applied separately for each of the spans that 170 occurred at a given time in a batch, whether or not the span was from the time when the masker noise 171 was present. First, individual spans were excluded if their RMS level was above a rejection criterion 172 that was twice as far from the peak as the regular criterion (i.e. we used two times the value of the 173 edge multiplier). This removed spans with particularly large-amplitude noises that would be rejected 174 no matter how low the noise was in any span they would be paired with. Spans that passed this first 175 criterion were paired by summing their waveforms point-by-point with one of the pair reversed in 176 polarity (to cancel the crosstalk) and from the summed waveform we calculated the reverse-pair-177 RMS value. Next, data from such a pair were excluded if the reverse-pair-RMS was above the 178 rejection criterion multiplied by the square-root of two to compensate for adding orthogonal noise 179 waveforms. The reverse-pair-RMS's of all the passed pairs in a batch were summed, and the sum was 180 divided by the number of spans that passed the rejection criteria. This yielded a single average RMS 181 value for the noise of a span in a given batch. This was done separately for successive spans across 182 the 14 epochs, yielding a time-course of RMS values across a trial. 183 The RMS values for each span in a batch were expressed in two ways: (1) RMS values were  184 converted to a linear version of dB SPL ("linear SPL") using the appropriate acoustic calibration. 185 These averages, converted to dB, were used when plotting the amplitudes in dB SPL.
(2) RMS values 186 were normalized by dividing each span by the average RMS value of the spans in epoch 10 of the 187 batch. For each method, data at each successive span were combined across batches by averaging the 188 RMS values. Batches were identified as "active" or "passive" and were averaged separately. In some 189 subjects, crosstalk sound from the highest pedestal levels was not canceled by averaging alternated-190 sign waveform pairs because the tone bursts randomly varied in amplitude so that adjacent 191 waveforms did not always have the same amplitude tone burst and therefore did not cancel. Data 192 from these pedestal levels were excluded from plots (31%, on average, including all of the 80 dB 193 pedestals); otherwise differences in pedestal level were ignored because we found no systematic 194 differences in ear-canal noise levels from batches with different tone burst pedestal levels. were not used to avoid times after MEM contractions would have decayed). If the increase in click 211 amplitude exceeded 0.2 dB, data from that trial were not used. With this criterion, data from ~0.5 to 212 4% of trials across subjects were excluded. However, because the rejected trials were not 213 systematically from certain subjects or pedestal levels, we think these rejected trials were not due to 214 actual MEM contractions. 215 The spectra of the ear-canal noise were obtained by a filter-bank method similar to that used by 216 Francis and Guinan (2010). We used zero-phase-change FIR digital filters. Individual filters were 217 500 Hz wide, with center frequencies 250 Hz apart (they overlapped), and extending from 500 to 218 4000 Hz. Span waveform pairs were combined with one of the pair reversed, so as to cancel any 219 crosstalk. They were accepted or rejected by their RMS values as described above, and then each 220 accepted pair was filtered to obtain its spectra. For each subject, span spectra were combined by 221 averaging in 6 groups: for epochs 1-9, all spectra were combined in a single group, and for epochs 10 222 to 14, all of the spectra from each epoch were combined into separate epoch averages. In all cases, 223 spectra from active and passive trials were combined separately. 224

Statistical analysis 225
To determine if changes in ear-canal sound recordings were statistically significant, we used a 226 bootstrap test (an ANOVA could not be used because the data were not normally distributed, see Fig.  227 2). Bootstrap tests were applied separately on each subject and each activity group (active or passive) 228 using averages of the span data in epochs 10 to 14 (each epoch averaged separately To compare whether the reduction in the ear-canal noise from epoch 10 to epochs 11-14 was more in 244 the active trials than in the passive trials of a subject, new pseudo-average values of the changes from 245 epoch 10 to epochs 11-14 were calculated separately for the active and passive trials as described 246 above. We calculated the noise reduction as: (epoch 10 -epoch X). From these new pseudo-247 averages, for each epoch we calculated the additional reduction of the ear-canal noise in the active 248 trials compared to the passive trials (i.e. the active value minus the passive value) and if this value 249 was less than zero, the comparison was scored as false. This was done 100,000 times and the fraction 250 false was taken as the probability that the hypothesis was false. This is the p value for the hypothesis 251 that the reduction of ear-canal noise from epoch 10 to epochs 11-14 was more in the active trials than 252 in the passive trials. 253 Results 254

No noise rejections 255
Ear-canal noise levels, expressed as dB SPL values in successive 25 ms time spans (Fig. 4A, B), 256 were measured when the subjects were doing the 2IFC task (active trials) and when subjects sat 257 quietly without doing the task (passive trials). The overall noise levels varied across subjects and 258 overlapped considerably. To make the trends easier to see, each set of data was normalized (using 259 SPLs as linear numbers) relative to their average value in the base epoch (epoch 10) and is replotted 260 in Figure 4C, D. In both active and passive trials the noise levels bounced around baselines that 261 remained relatively constant until the beginning of the epochs with masking noise, i.e. starting at 4 262 seconds in Figure 4. After the noise onset, the active and passive trials showed different behavior. In 263 the active trials, the noise level decreased near the time when the masking noise started (Fig. 4C). In 264 contrast, in the passive trials there was no clear trend (Fig. 4D). These data show there is a big 265 difference in the active versus the passive trials that first occurred when the subject had to attend to 266 doing the task. It shows that the overall noise level was strongly influenced by whether the subject 267 was doing the task, or not. This difference is present in the data without any data processing. 268 However, it is well known that subject movement can produce noise that is picked up by an ear-canal 269 microphone, and that subjects never sit completely still. Thus, a hypothesis that may account for 270 these data is that the subject sat more still when paying attention to doing the task. 271

Strict Noise Rejections 272
In almost all experiments in which OAEs are measured, an artifact rejection system is used in which 273 the experimenter chooses a sound level criterion and portions of the recording above this criterion are 274 removed from consideration. We used an artifact rejection system with the criteria varied by setting 275 different "edge-multiplier" values (see Methods). For an edge-multiplier of 2, figure 5 shows 276 example plots versus time of both the ear-canal noise and the fraction of spans rejected, for both 277 active and passive trials. An edge-multiplier of 2 provides a strict cut off that removes all spans with 278 RMS values above the peak region in histograms of RMS values (see Fig. 2). 279 After the rejection of spans with high noise levels by applying an edge-multiplier of 2, each subject's 280 average noise level was relatively constant during the time before the masker noise began (Fig. 5A,  281 B). The different SPL values for the ear-canal noise of different subjects are presumably due, at least 282 in part, to differences in ear-canal volumes and the depths of insertion of the probes. In both active 283 and passive trials (Fig. 5A & B), two subjects (323 and 326) had visible reductions in the overall dB 284 SPL level of the ear-canal noise when the task-ear masker was on. These reduction are more easily 285 seen in Figure 5C and D, which show the same data normalized to its value in epoch 10. The time 286 courses of the decreases in ear-canal noise in these two subjects (323, 326) are similar to the time 287 courses expected from MOC inhibition elicited by the task-ear masking noise (Fig. 5E). These two 288 subjects were the only ones with SOAEs. A hypothesis that fits these data is that in these two 289 subjects, the ear-canal "noise" was partly due to SOAEs that were inhibited by MOC activity elicited 290 by the task-ear masker. 291 In the passive trials, after applying an edge multiplier of 2 to remove bursts of noise, the changes in 292 ear-canal noise from epoch 10 to epoch 13 were all small, but some were statistically significant. The 293 largest changes were in subjects 323 and 326 who had decreases of 3.9% and 2.4%, respectively, that 294 were highly statistically significant (p<0.0001). In three other subjects, there were statistically 295 significant decreases of 0.3%, 0.3% and 0.8% (p=0.016 for the least significant of these). The very 296 small decreases in these three subjects may be due to MOC inhibition of un-noticed SOAEs or other 297 ear-canal noise, but their time courses are too poorly defined to help substantiate this. In one subject, 298 there was a decrease of 0.02% that was not statistically significant (p=0.47). In the two remaining 299 subjects there were small increases: one increase was 0.19% but not significant (p=0.14), the other 300 (subject 319) was an increase of 0.45% and was statistically-significant (p=0.0002). 301 In the active trials, after applying an edge-multiplier of 2 to remove bursts of noise, all of the subjects 302 had decreases in ear-canal noise from epoch 10 to epoch 13 that were statistically-significant 303 (p=0.00016 for the least significant). For subjects 323 and 326 the decreases were 5.4% and 3.2%, 304 respectively, and for the six other subjects the decreases averaged 1.9% (range 0.27% to 2.6%). The 305 largest decreases (in subjects 323 and 326) had time courses consistent with most, or all, of the 306 decrease being from MOC inhibition elicited by the masker noise. The time courses of the decreases 307 in the other six subjects are difficult to see in Figure 5C. To make these time courses more visible, 308 we adjusted the magnification and offset of each so that their average in epoch 13 was zero while the 309 average in epoch 10 was kept equal to 1. The result ( Figure 5F) shows the degree to which the time 310 courses of the reductions in ear-canal noise were similar across these six subjects. The time course of 311 these reductions appears to have a slightly slower onset than the larger reductions seen in subjects 312 9 323 and 326 (Fig. 5C, D vs. F). However, the waveforms are somewhat noisy and the differences 313 between them are not particularly clear. 314 We compared the decrease in ear-canal noise from epoch 10 to epoch 13 in active versus passive 315 trials for an edge multiplier of 2. In 7 of 8 subjects the percentage decrease in ear-canal noise was 316 more for the active trials than for the passive trials. The active change minus the passive change 317 averaged 1.04%, range -0.03% to +2.5%). The greater decreases in the active trials were statistically 318 significant in 6 of the subjects (largest p=0.005) and the one increase was not significant (p=0.56), 319 In addition to measuring the changes in ear-canal noise, we also measured the fraction of spans that 320 were rejected. For an edge-multiplier of 2, the fraction of spans rejected are shown in Figure 5G, H. 321 Near the end of the trials, when the subject had to do the 2IFC task, there was a clear difference in the 322 fraction of spans rejected in active versus passive trials. In active trials the fraction rejected went 323 down shortly after the masker noise started, whereas in passive trials the fraction rejected was little 324 changed or went up (Fig. 5G, H). For active trials, all subjects had a decrease in the fraction rejected 325 from epoch 10 to epoch 13 (average decrease = 0.107 range 0.014 to 0.23). Five of these were 326 statistically significant (highest p=0.045) and three were not. In contrast, none of the passive trials 327 had a statistically significant change (at the 0.05 level) in the fraction rejected in either direction over 328 these same intervals. 329 Both the fractions rejected and ear-canal noise levels show the pattern over time of the bursts of noise 330 that were present in the original data. The data of Figure 5G show that subjects reduced their 331 production of large bursts of noise when doing the task. A hypothesis that fits these data is that large 332 bursts of ear-canal noise are due to subject movements. With this hypothesis, the time courses of the 333 decreases in the large-amplitude noises in Figures 4 and 5 shows the time courses over which 334 subjects decreased movements as they directed their attention to doing the 2IFC task. In contrast, the 335 large amplitude noises were little changed in the passive trials. 336

Varying noise rejections 337
The data of Figure 5 were for a strict noise-rejection criterion: an edge-multiplier of 2. For edge 338 multipliers from 2 to 100, the reductions in ear-canal noise and the fraction of spans rejected for the 339 active trials of all subjects are shown in Figure 6. Higher edge-multipliers reject fewer spans, but the 340 pattern across time of the fraction of spans rejected changed little as the edge-multiplier was changed. 341 When the criterion removed only very highest level sounds (edge multiplier of 100), or when the 342 criterion removed all of the noise levels above the main peak in the span RMS histograms (edge 343 multiplier of 2), the fraction rejected was lowest at the time when the subject had to make the 2IFC 344 judgment (Fig. 6). Further, for each subject, the time courses of the reductions in ear-canal noise 345 were very similar to the time courses of the fractions rejected, presumably because both were due to 346 the same underlying cause. 347 The changes in ear-canal noise as the edge multiplier was changed from zero to 100, quantified as the 348 change from epoch 10 to epoch 13, are shown in Figure 7. An edge-multiplier of zero applies a noise 349 rejection criterion at the peak of the histogram of span RMS levels (see Fig. 2). Also included in 350 Figure 7 are the changes from epoch 10 to epoch 13 of the raw data (data with no noise rejection 351 applied). As the edge-multiplier was made less strict (i.e. had higher values) and fewer spans were 352 rejected, the changes between epoch 10, and epoch 13 became larger for all subjects in active trials, 353 but remained small in passive trials (Fig. 7A, B). To determine the extent to which the ear-canal 354 noise was reduced more in active trials than passive trials, the difference between the two conditions 355 is shown in Figure 7C. The difference was large when the edge multiplier was high and removed 356 only the highest-level noise bursts, but as the edge multiplier was made more strict, the difference 357 between the active and passive trials became less and less (Fig. 7C). For edge multipliers less than 1 358 there was almost no additional decrease (the decreases were less than 1%) in ear-canal noise 359 produced in the active trials compared to the passive trials (Fig. 7C, inset). Note that using severe 360 criteria (edge multipliers of 1 or less) did not remove the ability to see the small reductions in ear-361 canal noise in subjects 323 and 326 (Fig. 7A, B) -reductions that we attribute to the masking noise 362 evoking MOC activity that reduced SOAEs and other noise of cochlear origin in these two subjects 363 (Fig. 7B). 364

Noise spectra 365
Although the overall noise levels varied across subjects, all subjects showed similar patterns of ear-366 canal noise as a function of frequency. The noise amplitudes were largest at the lowest frequencies, 367 were smallest at mid frequencies (2-3 kHz) and increased at higher frequencies (solid lines in Fig.  368 8A, B). The decrease from the original spectra to the spectra after applying an edge-extender of 2 was 369 greater as frequency decreased (dashed lines in Fig. 8A, B). After noise bursts were removed by 370 applying an edge multiplier of 2, there was little change in ear-canal noise from epoch 10 to epoch 13 371 at most frequencies (Fig. 8C, D). However, in the two subjects who showed reductions in SOAEs 372 and/or other ear-canal noise with a time course appropriate for a MOC inhibition (subjects 323 and 373 326), there were decreases in the 1 to 2 kHz range (Fig. 8C, D). This frequency range approximately 374 corresponds to the frequencies of these subjects' SOAEs (Fig. 8E, F) and is also consistent with these 375 changes being due to MOC inhibition. 376 Discussion 377 During the behavioral task we found reductions in ear-canal acoustic noise that were very large when 378 no noise bursts were rejected, but became small when a strict criterion was used that removed most 379 of the bursts of noise. The largest reductions in ear-canal noise were for active trials. We attribute the 380 reductions in ear-canal noise as being due to two main sources: (1) inhibition from MOC efferent 381 activity elicited by the task-ear masker noise, and (2) a reduction in subject motion concurrent with 382 the subject attending to the task. 383

Reduction of ear-canal noise from MOC inhibition elicited by contralateral sound 384
A standard way of measuring MOC inhibition on OAE responses in one ear (here called the 385 ipsilateral ear) has been to elicit MOC activity by contralateral acoustic stimulation (CAS). In the 386 passive trials we did a measurement like that with the CAS being the task-ear masker. One difference 387 from a typical MOC-effect measurement was that instead of measuring the effect on a sound-evoked 388 OAE, we measured the effect on ear-canal acoustic "noise" (i.e. sound within the ear canal that was 389 not evoked by a presented sound). In two subjects (323 and 326) we found strong evidence for 390 reductions in ear-canal noise produced by CAS-elicited MOC inhibition: (1) the time courses of the 391 reductions followed the typical time course of MOC inhibition produced by contralateral sound (Fig.  392 5), (2) as the criteria for removing ear-canal noise were made more strict, the changes from epoch 10 393 to epoch 13 did not go away, consistent with these changes not being due to changes in subject 394 motion (Fig. 7A, B), and (3) the changes were found in both passive and active trials (Fig. 5C, D). 395 These data fit with the hypothesis that in these two subjects, some of the ear-canal noise originated in 396 the cochlea, and that MOC activity elicited by the masker CAS reduced cochlear amplifier gain 397 thereby reducing the ear-canal noise. These two subjects were also the only subjects who had 398 SOAEs, and it seems likely that much, or all, of the change was due to MOC inhibition of SOAEs 399 (Mott et al., 1989;Harrison and Burns, 1993;Zhao and Dhar, 2010). However, it is also possible that 400 some fraction of the change was actually MOC inhibition of a random signal that originated within 401 the cochlea. Consistent with the hypothesis that some ear-canal noise in humans originates in the 402 cochlea, Nuttall et al. (1997) found that basilar membrane velocity noise was enhanced by cochlear 403 amplification and inhibited by MOC stimulation. This basilar membrane velocity noise can be 404 expected to create backward-traveling noise waves that produce noise in the ear canal. 405 In addition to the two subjects with easily-seen decreases in ear-canal noise in passive trials, three 406 other subjects also had very small, but statistically-significant decreases in ear-canal noise from 407 epoch 10 to epoch 13. These may also have been MOC inhibitions of ear-canal noise or SOAEs that 408 were too small to see. Overall, our finding of little or no CAS-elicited reduction in the ear-canal noise 409 in subjects with no SOAEs is consistent with the hypothesis that in subjects with no SOAEs there is 410 little or no noise in the ear canal that originated from within the cochlea. 411 The data without any noise rejection (Fig. 4) provide clear evidence that subjects reduced their ear-412 canal noise at the time the task was done. Several lines of evidence indicate that this was caused 413 mostly by reduced subject motion, and not by task-elicited MOC activity reducing noise that 414 originated within the cochlea. First, the largest noise bursts seem highly likely to have been produced 415 by subject motion because their amplitudes are too high to be accounted for by any known cochlear 416 mechanism. This is consistent with the normal interpretation in OAE measurements that large bursts 417 of noise are due to subject motion. Second, when a strict criterion for removing large-amplitude 418 noises was applied (e.g. an edge multiplier of 2 or less) there was almost no additional reduction in 419 ear-canal noise in the active trials compared to the passive trials (Fig. 7C). Finally, one might think 420 that attention-elicited MOC activity that reduced ear-canal noise would lead to a reduction of the 421 number of spans rejected at that time. This explanation might then account for the pattern in Figure 6  422 where the reductions in the ear-canal noise and in the number of spans rejected have similar time 423 courses. However, this explanation doesn't fit with there being big reductions in ear-canal noise 424 when the noise cut-off criterion was high (large edge multipliers), and small reductions as the cut-off 425 criterion was made stricter. This pattern implies that when the subject did the task, the largest noise 426 bursts were reduced more than the smallest noise bursts. For this pattern to be produced by MOC 427 inhibition, the largest noise bursts would have to be inhibited more than the smaller noise bursts, 428 which is opposite the pattern actually found for MOC inhibition at these sound levels (Guinan and 429 Gifford, 1988; Guinan and Stankovic, 1996;Cooper and Guinan, 2006;Bhagat and Carter, 2010). 430 Thus, the hypothesis that attention reduces ear-canal noise through MOC inhibition doesn't fit the 431 data for most subjects. A hypothesis that fits the data more broadly is that when attending to the task, 432 the subjects sat more still and generated fewer bursts of noise. 433 It is interesting that the two subjects who showed clear evidence for CAS-elicited MOC inhibition of 434 ear canal noise (323 & 326) also had slightly more change from epoch 10 to epoch 13 in active 435 compared to passive trials (~1-2% greater during active trials; Fig. 7C, inset). One interpretation of 436 this is that in these two subjects, task-related attention slightly increased the MOC activity and 437 thereby produced a slightly greater epoch 10 to epoch 13 change in the active trials. However, since 438 these changes were so small and absent in 6/8 subjects, we do not conclude that attention reduces ear-439 canal noise through MOC inhibition. 440

Comparison with previous reports 441
De Boer and Thornton (2007) reported reductions in ear-canal noise level when subjects did an 442 auditory task or paid attention to a movie. They interpreted the changes they found in ear-canal noise 443 as due to changes in subject-generated noise that were affected by attention and were also affected by 444 whether the subject noise interfered with performance of the task. Their interpretation is consistent 445 with ours. 446 In contrast, Walsh et al. (2014;2015) reported a large decrease (~3 dB) in ear-canal noise in all of 447 their subjects when the subject did a behavioral discrimination compared to during passive listening. 448 They interpreted the decrease as being produced by MOC inhibition of ear-canal noise. The 449 interpretation that this change was due to MOC inhibition is questionable for several reasons. A 3 dB 450 reduction is at the high end of typical MOC effects on OAEs    (including noise  488 and with no artifact rejection) during the experiment (as in Fig. 4)  when the 2IFC target tones were presented (compare Figs. 1 and 6). The time course of the reduction 499 in ear-canal noise shows that the buildup and decay of attention occurred over several hundreds of 500 ms. 501 While we did not find that attention changed ear-canal noise through MOC inhibition, we did find 502 that the decrease in ear-canal noise was a very robust indicator of whether subjects were or were not Recently, Gruters et al (2018) found an interaction between saccadic eye movement and changes in 511 ear-canal sound pressure that lasted for 10's of milliseconds. The infrasounds produced by such eye 512 movements would have been filtered out in our measurements, but they do point out that there are 513 many subject motion changes that may affect ear-canal noise. In addition, Braga et al (2016) found 514 that saccade rates decrease during auditory attention. Thus, it is possible that as subjects attended to 515 the auditory task, saccadic eye movements settled down, and this has a role in reducing ear-canal 516 noise. If true, this hypothesis would indicate that eye-tracking might also help to sort out the origin of 517 changes in ear-canal noise during task performance. 518 Our results indicate that before making definitive conclusions about the origin of changes in OAEs or 519 ear-canal noise measured during a behavioral task, it is necessary to take into account all other 520 sources that may affect ear-canal sound levels. This is especially true when studying MOC efferent 521 effects, since extremely subtle motion artifacts may closely resemble MOC effects yet not be related 522 to MOC inhibition. 523

Conflict of Interest 524
The authors declare no competing financial interests. presented. In the task ear, stimuli were presented in 400 ms "epochs" with 50 dB pSPL clicks 641 presented every 25 ms throughout. There were 1-10 (randomly chosen) epochs before three epochs 642 with 50 dB SPL masking noise, followed by one epoch at the end. The epochs from "base" to "reply" 643 were present on every trial. The last two epochs with noise (epochs 12&13) also contained tone pips, 644 which were the same amplitude in the no-task trials, and different amplitude in task trials. At the end 645 of each trial during the task, subject had to push a button to indicate which tone pip was louder.