Auditory Azimuthal Localization Performance in Water as a Function of Prior Exposure

Introduction

This study assessed auditory azimuthal localization performance in water for 45 normal-hearing adults with various durations of prior exposure to water.

Sound localization in humans mainly depends on three cues: interaural time differences (ITDs), interaural level differences (ILDs), and spectral cues (for a review, see Moore, 2003). ILDs and spectral cues are extracted from the high-frequency spectrum of sounds. Binaural cues (i.e., ITDs and ILDs) allow left/right discrimination and azimuth resolution. Spectral cues allow front/back and up/down discriminations and elevation resolution. The set of direction-dependent cues is transformed by the auditory system into spatial coordinates. This complex audio-spatial map depends on anatomical characteristics and develops through experience with visual and vestibular feedback. Spatial hearing capacities of adult animals are substantially malleable by learning (King et al., 2001; Knudsen, Esterly, & Olsen, 1994). In humans, the localization abilities of children are similar to those of adults in spite of the change in morphology during growth (Clifton, Clarkson, Gwiazda, Bauer, & Held, 1988). As detailed further in the following, studies of learning (i.e., of behavioral plasticity) in human adults may yield practical implications for numerous populations (e.g., hearing-impaired listeners with reduced sensitivity to spatial cues, normal-hearing listeners with age-related decline in binaural processing, and normal-hearing users of virtual auditory displays).

The adaptability of the mature auditory system in spatial hearing has been intensively assessed in behavioral studies, under either “active” or “passive” exposure to altered cues. In studies of active exposure, artificial alterations in the relationship between source direction and perceived cue were administered during testing sessions in which the participants were provided with feedback as to the correct target direction (Abel & Paik, 2004; Butler, 1987; Hofman, Vlaming, Termeer, & van Opstal, 2002; Musicant & Butler, 1980; Russel, 1976; Shinn-Cunningham, Durlach, & Held, 1998; Young, 1928; Zahorik, Bangayan, Sundareswaran, Wang, & Tam, 2006). In studies of passive exposure, participants were asked to wear a system distorting the relationship between source direction and perceived cue over several hours or days during waking hours. Localization abilities were measured at regular intervals without any feedback information during measurements or control of auditory inputs during daily activities (Held, 1955; Hofman, van Riswick, & van Opstal, 1998; Javer & Schwarz, 1995; van Wanrooij & van Opstal, 2005). In both sets of studies, localization performance initially diminished compared to normal-listening conditions but started to improve within the first hours of exposure. Improvement was faster and more complete for spectral than for binaural distorted cues.

In water, spectral cues and ILDs are theoretically lost mainly because of the impedance mismatches between water and upper body (i.e., pinna cavities, ear canal, ear cartilage, inner ear, and head bones; for details, see Savel, Drake, & Rabau, 2009). Moreover, the propagation velocity of sound waves in seawater (1,500 m/s) is about 4.5 times greater than in air (343 m/s), so ITDs are considerably reduced. In other words, vertical localization performance should be poor in water and azimuthal performance should solely rely on smaller than normal ITDs. (To our knowledge, no study except Ross, Crickman, Sills, & Owen, 1969, assessed vertical localization or orientation in water. This would be of interest for diver’s both orientation and localization of various sources, e.g., other divers below, boats above. We performed informal measurements of front/back discrimination using broadband noise in 10 experienced divers and observed poor discrimination.) The size of the ITD as a function of target azimuth can be estimated using Equation 1:

I T D = ( r × rad θ + r × sin ( rad θ ) ) × t ,

where r is the radius of the head (9 cm), θ is the target angle (azimuth, in degrees), and t is the time needed to travel 1 cm. ITDs for target angles ranging from 0° (median axis) to 90° (ears axis) thus vary from 0 µs to 680 µs in air but to 155 µs in seawater. Left/right discrimination should remain possible in water because the ITDs produced for target angles of 10° or beyond are above the ITD at threshold for accurate left/right discrimination (Klumpp & Eady, 1956; McFadden & Pasanen, 1976; Wright & Fitzgerald, 2001; Yost, 1977). In concordance, several past studies reported correct left/right discrimination in water (Feinstein, 1973; Hollien 1973; Norman, Phelps, & Wightman, 1971). However, if listeners associate ITDs in water with their “normal” (air) directions, they should exhibit a directional bias toward the center of the horizontal plane. A small medial bias was reported in experienced scuba divers (Wells & Ross, 1980). The authors suggested that the bias was undersized compared to that predicted by the reduction of ITDs because the participants would have adapted during their prior exposures to water. This would correspond to the previously described passive exposure to altered cues. The hypothesis formulated by Wells and Ross (1980) has not been verified yet: Past studies in water exclusively involved experienced divers, with long prior exposure. If adaptability exists in water, the resulting modification of the audio-spatial map should occur over exposure time in the same way as it has been found to occur in air.

Research Objectives and Hypotheses

The aim of the study was to assess in water the possible dependency of auditory azimuthal localization performance on overall duration of exposure to the environment. A transversal study measuring performance in the same participants over several hours of immersion in water would have been unachievable. Thus, the relationship between performance and exposure was examined in a single session within large cohorts of participants. The cohorts were composed of normal-hearing adults with various durations of prior exposure to water but with no experience in psychoacoustical tasks. Exposure duration was defined as the sum of dives performed prior to and independently of the study.

Performance in water was expected to be exclusively based on smaller than normal but detectable ITDs. The following results were thus awaited regardless of exposure duration: poor capacity of front/back discrimination, high capacity of left/right discrimination, and medial bias in judgments about the target azimuthal direction. Moreover, a negative correlation was expected between medial bias size and prior exposure duration, as a result of adaptation.

We used low-frequency sounds (i.e., frequency <500 Hz) as auditory targets for two reasons. First, both hearing sensitivity (Brandt & Hollien, 1967; Hollien & Feinstein, 1975) and spatial hearing (Feinstein, 1973; Hollien, 1973; Norman et al., 1971) are better in water with low-frequency stimuli. Second, we aimed at excluding the involvement of spectral cues and ILDs (although both cues are theoretically lost in water), available in air at high frequencies. As poor front/back discrimination was awaited but head movements improve front/back discrimination (in air: Thurlow & Runge, 1967; Wightman & Kistler, 1999; in water: Savel et al., 2009), head motion was not allowed.

Method

Apparatus and Stimuli

The localization tests were conducted in open sea (temperature = 19°–23° C) and controlled by a PC on board a boat. The apparatus was located 6,000 feet from the coast, 10 feet under the sea surface, and 50 feet above the seafloor. In Experiment 1 the eight loudspeakers (Aquamusique HP 062) were positioned at different azimuths (see Figure 1) on a cage with acoustic-transparency properties. Participants were seated in the center of the cage, 3.3 feet from each loudspeaker. In Experiment 2 the loudspeaker/head distance was increased to 65.6 feet, which avoided visual cues and proximity effects (Brungart, Durlach, & Rabinowitz, 1999). There was no cage and participants stood erect. The azimuths of the loudspeakers were the same as in Experiment 1.

OPEN IN VIEWER

Figure 1.

This illustration shows the loudspeakers and participant positions for the localization tests. The head/loudspeaker distance (d) was 3.3 feet in Experiment 1 and 65.6 feet in Experiment 2.

The signals were generated digitally at a 20-kHz sampling rate using a digital array processing card and a digital-to-analog converter (DAC). The output of the DAC was amplified and routed to one of the eight loudspeakers via one of eight channels of a divider. The target was a train of 300 ms bursts, including 25 ms cosine rise/decay times. The repetition rate of the bursts was 2/s and the overall duration of the train was 2.7 seconds. In Experiment 1 the bursts were 400-Hz pure tones. In Experiment 2 the bursts were filtered noise bands (20–400 Hz; center frequency = 210 Hz) whose spectral characteristics were chosen so as to resemble the typical noise produced by a motor boat moving at average speed. The stimulus level at the theoretical center of the head was 100 dB SPL in Experiment 1 and 90 dB SPL in Experiment 2. The lower SPL in Experiment 2 was a consequence of the greater distance. The stimulus SPLs were about 60 to 70 dB above hearing threshold in water for 250 to 500 Hz tones, were thus comfortably audible while being below discomfort and pain thresholds (in water, hearing, discomfort, and pain thresholds are 45–60 dB SPL above those in air; for measurements in water of hearing threshold and aversion to sound and for explanations about how hearing mechanisms in water drastically decrease hearing sensitivity, see, e.g., Brandt & Hollien, 1967; Sims, Fothergill, & Curley, 1999; Wainwright, 1958) and causing no risk of overexposure.

The output levels of the loudspeakers were carefully controlled and matched using the same procedure as that described in Savel and colleagues (2009, see footnote 4). No loudness or timbre differences between loudspeakers were reported by three experienced listeners. The ambient noise was continuous, had maximum energy below 1000 Hz, and had an average overall level of 65 dB SPL. The target-to-noise ratios were well above those at which broadband background noise affects localization accuracy in air (e.g., Jacobsen, 1976; Lorenzi, Gatehouse, & Lever, 1999). The targets were clearly audible for all participants. The tests were interrupted when substantial waves or boat courses caused the noise level to increase by more than 10 dB SPL.

Data Analysis

We assessed within-listener variability in Experiment 1. Each response was provided with a button number and variability across 48 responses was quantified using coefficients of variation (CV in % = standard deviation/mean × 100). This allowed verifying whether participants used the available response choices independently of their experience in scuba diving.

In both experiments we assessed individual front/back discrimination ratios (i.e., ratio of “true-hemifield” responses – ratio of “false-hemifield” responses; see Figure 2a) and individual left/right discrimination ratios (i.e., ratio of “true-side” responses – ratio of “false-side” responses; see Figure 2b). In Experiment 1 individual raw amounts of “true” (hemifield or side) responses were compared to amounts of “false” responses using McNemar’s χ² tests. Discrimination was considered as accurate if both the χ² test was significant (p < .05) and the discrimination ratio had positive sign (see asterisks in Table 1). In Experiment 2 the amount of trials was too low for the χ² test to be significant. We thus considered discrimination as accurate for ratios amounting to +0.8 or above (see asterisks in Table 2).

OPEN IN VIEWER

Figure 2.

Details on the (a) calculation of front/back discrimination ratios and (b) left/right discrimination ratios.

OPEN IN VIEWER

Table 1:

Individual Data (17 participants) for Each Variable Under Study in Experiment 1

Participant	Exposure Duration (hours)	Exposure Duration (log)	Within-Listener Coefficient of Variation in Responses (%)	Front/Back Discrimination Ratio	Left/Right Discrimination Ratio	Signed Error (°)
1	0.5	−0.3	29	−0.1	0.3*	−28
2	1	0.0	43	0.3	0.4*	−39
3	2	0.3	46	0.2	0.8*	−13
4	3	0.5	47	0.5 *	0.1	−44
5	4	0.6	46	−0.0	0.4*	−35
6	15	1.2	63	0.1	0.3*	−34
7	25	1.4	78	0.2	0.4*	−39
8	25	1.4	51	0.0	0.3	−23
9	25	1.4	52	0.2	0.9*	−15
10	30	1.5	61	0.1	0.4*	−25
11	40	1.6	45	0.2	0.9*	−6
12	80	1.9	52	0.4 *	1.0*	−3
13	108	2.0	38	0.1	0.5*	−29
14	135	2.1	52	0.2	0.4*	−17
15	185	2.3	56	0.3	0.8*	−16
16	225	2.4	43	0.2	0.7*	−6
17	500	2.7	62	0.2	0.7*	−24
Median	25	1.4	51	0.2	0.4	−24
Interquartile range	104	1.4	10	0.1	0.4	18

Note. Outlier values are underlined. Asterisks indicate “accurate” discrimination (see text for details).

OPEN IN VIEWER

Table 2:

Individual Data (28 participants) for Each Variable Under Study in Experiment 2

Participant	Exposure Duration (hours)	Exposure Duration (log)	Front/Back Discrimination Ratio	Left/Right Discrimination Ratio	Signed Error (°)
18	3	0.5	−0.3	0.0	−35
19	4	0.6	0.8*	0.5	−25
20	5	0.7	0.2	0.8*	−22
21	6	0.8	0.5	1.0*	−30
22	7	0.9	−0.3	–0.2	−20
23	7	0.9	0.0	0.8*	−28
24	8	0.9	−0.5	0.7*	−23
25	8	0.9	−0.5	0.3	−39
26	8	0.9	0.2	0.3	−9
27	9	1.0	0.3	0.7*	−31
28	10	1.0	0.0	0.2	−35
29	10	1.0	0.0	1.0*	−22
30	10	1.0	−0.3	0.8*	−35
31	11	1.0	0.5	0.8*	−6
32	12	1.1	−0.3	0.3	−49
33	13	1.1	−0.5	0.7*	−25
34	14	1.2	−0.3	0.7*	−24
35	30	1.5	0.0	0.7*	−16
36	35	1.5	0.8*	0.8*	−23
37	40	1.6	0.3	1.0*	−9
38	50	1.7	0.5	1.0*	3
39	50	1.7	−0.2	0.8*	−15
40	51	1.7	0.5	0.8*	−23
41	100	2.0	−0.2	1.0*	−10
42	245	2.4	−0.2	0.7*	0
43	300	2.5	0.8*	1.0*	2
44	500	2.7	0.5	0.7	−18
45	500	2.7	0.3	0.7*	0
Median	12	1.1	0.0	0.7	−22
Interquartile range	42	0.8	0.8	0.2	18

Note. Outlier values are underlined. Asterisks indicate “accurate” discrimination (see text for details).

To quantify directional biases for the six loudspeakers off the median axis (i.e., for all loudspeakers except 1 and 5; see Figure 1), side errors were excluded and hemifield errors were corrected (i.e., response and target angles in the rear transformed into [180 − x_i]). Individual signed localization errors were then computed for each trial (i.e., [unsigned response angle − unsigned target angle] in degrees) and averaged across loudspeakers and trials.

Two theoretical signed errors were estimated. The first was the error that should be obtained if the participant exhibited no particular bias, had no idea about the target direction, and responded randomly (= equal use of response possibilities). The second was the systematic medial bias that should be observed in case of no adaptation to the environment (= systematic underestimation of the target angle). This bias was assessed as follows: (θ_A − θ_W), where θ_W is the target angle in water and θ_A is the angle in air that produces the same ITD as θ_W according to Equation 1. Averaged across loudspeakers and trials, the random error amounted to −15° (95% confidence interval = 13 in Experiment 1, 27 in Experiment 2) and the medial-bias error amounted to −48° (confidence interval = 6 in Experiment 1, 16 in Experiment 2).

According to Shapiro-Wilk’s tests, several variables (i.e., exposure durations in both experiments, front/back and left/right discrimination ratios in Experiment 2) were not normally distributed (p < .05). Exposure durations were thus log-transformed. The correlations between log exposure duration and each variable were assessed using Spearman’s correlation coefficients (r_s) after exclusion of outlier values (i.e., values below [first quartile – (1.5 × Interquartile Range)] and above [third quartile + (1.5 × Interquartile Range)]; see underlined values in Table 1 and 2). Individual results are listed in Table 1 (Experiment 1) and Table 2 (Experiment 2).

Results and Discussion

Within-listener CVs in Experiment 1 ranged from 29% to 78% but were not correlated with exposure (n = 15, r_s = 0.25, p = .37).

Front/back discrimination ratios ranged from −0.1 to 0.5 (Experiment 1) and from −0.5 to 0.8 (Experiment 2). Their median values were low (Experiment 1: 0.2; Experiment 2: 0.0). Only two participants in Experiment 1 and three in Experiment 2 showed accurate front/back discrimination. False-hemifield responses represented 35% of the total amount of responses in Experiment 1 and 50% in Experiment 2. We observed no correlation between front/back discrimination and exposure (Experiment 1: n = 15, r_s = 0.42, p = .12; Experiment 2: n = 28, r_s = 0.26, p = .19).

Left/right discrimination ratios ranged from 0.1 to 1.0 (Experiment 1) and from −0.2 to 1.0 (Experiment 2). Their median values were fairly high (Experiment 1: 0.4; Experiment 2: 0.7). Fifteen participants in Experiment 1 and 20 in Experiment 2 demonstrated accurate left/right discrimination. False-side responses represented only 8% of the responses in Experiment 1 and 13% in Experiment 2. We found no correlation between left/right discrimination and exposure (Experiment 1: n = 17, r_s = 0.45, p = .07; Experiment 2: n = 25, r_s = 0.19, p = .37).

All 17 participants from Experiment 1 and 24 participants from Experiment 2 showed localization errors with negative signs. The error magnitudes widely varied across participants. The correlation between signed error and exposure was significant in both experiments (Experiment 1: n = 17, r_s = 0.54, p = .03; Experiment 2: n = 28, r_s = 0.59, p = .001). In Figure 3, the fitting functions (dashed lines) indicate that the errors obtained at about 0 hours exposure were closer to the bias predicted in case of no adaptation (open squares) than to the error predicted in case of random response about the target direction (open circles).

OPEN IN VIEWER

Figure 3.

Individual signed errors are plotted as a function of exposure duration for Experiment 1 (left panel) and 2 (right panel). Exposure values are reported in both log transform (plain) and hours (italics). The horizontal lines at the top of the panels indicate the error that would be observed if judgments about the target direction were perfectly accurate (signed error = 0°). The open circles show the error that would be observed if the participants responded randomly. The open squares show the bias that would be observed if no adaptation to the environment had occurred. Bars represent  ±95% confidence intervals. The correlations between observed signed error and log-transformed exposure were assessed using Spearman’s correlation coefficients (bottom right corners) and represented using linear fitting functions (dashed lines).

In summary, the ability to discriminate front from back was generally poor, the capacity to discriminate left from right was mostly accurate, and a medial bias was frequently observed in responses about the target direction, regardless of prior immersion in water. This likely indicates that smaller-than-normal but above-threshold binaural cues were used. The size of the bias decreased as exposure duration increased, which may reflect adaptation/plasticity in water resulting from passive exposure.

It could be argued that the participants with low exposure (i.e., with poor experience in diving) exhibited poorer performance mainly because they felt uncomfortable in water. If this were true, these participants would possibly have exhibited three behaviors related to discomfort: greater air consumption throughout the experiment, atypical within-listener variability in Experiment 1, and inability to reach the expected pointing accuracy in the procedural training of Experiment 2. Those behaviors were not observed. Moreover, uncomfortable participants would have demonstrated uniformly poor front/back and left/right discriminations, no particular directional bias, and/or large signed errors. This could possibly be the case for three participants in Experiment 2 (i.e., participants 22, 26, and 28; see Table 2). When they were excluded from the cohort, the patterns of correlation previously described were not altered and were even reinforced (front/back discrimination: n = 25, r_s = 0.26, p = .21; left/right discrimination: n = 24, r_s = 0.13, p = .54; signed error: n = 25 r_s = 0.69, p = .0002). Thus, the relationship between response direction and prior exposure likely reflects the time course of the adaptability to altered binaural cues. The large across-participant variability found at particular values of exposure duration (see Figure 3) could reflect variability in the auditory inputs individually received during passive exposure.

In a past study in air, an artificial ITD of 342 µs was presented to the participants over several days (Javer & Schwarz, 1995). A bias of 42° toward the periphery of the horizontal plane—thus providing the theoretical signed error with opposite sign but similar size compared to that involved here—was predicted. Localization responses to a target presented at 0° were regularly measured in the same six participants from 0 to 130 hours of exposure. The mean signed error amounted to about 0° before exposure (i.e., with normal cues) and varied from 30° (h = 0) to 9° (h = 130) during exposure. We attempted to fit the data from both the present and past studies using the exponential-decay function in Equation 2, where x(d) is the signed error at exposure duration d (in hours), A is the asymptotic error, L is the total decrease in error (i.e., x[0] – A), and k is a constant.

x ( d ) = L exp – k d + A

To allow between-study comparison, the error signs in the present study were inverted. Fitting was poor for Experiment 1 but was plausible for both Experiment 2 (25 participants included, 3 “uncomfortable” excluded; r² = .53) and the study by Javer and Schwartz (mean data; r² = 0.87). In Figure 4, the fits indicate substantial but incomplete audio-spatial remapping (value of A ± 95% confidence interval = 8–16 in Javer & Schwarz, 1995, –10–18 in Experiment 2). In Javer and Schwarz (1995), the experiment stopped at 130 hours of exposure, where one half of the participants showed errors at/below 0°. Thus, further decrease in error could have been occulted. This would explain why the estimate of A in their study is above that in the present work (see rightmost part of the fits in Figure 4).

OPEN IN VIEWER

Figure 4.

The exponential-decay function (dashed lines) of the form indicated in Equation 2 was applied to both the individual data from Experiment 2 (filled triangles) and the mean data from a past study in air (Javer & Schwartz, 1995; large open triangles). In the present study, the acoustic properties of seawater were expected to produce a bias of 48° toward the center of the horizontal plane (filled square). In the past study, the presentation of a 342-µs artificial interaural time difference (ITD) was expected to produce a bias of 42° toward the periphery (open square). To allow comparison, the error signs in the present study were inverted.

Previous studies of exposure to either spectral (e.g., Hofman et al., 1998) or binaural (e.g., Javer & Schwarz, 1995) altered cues reported little or no aftereffect. Namely, the worsening of localization abilities compared to normal listening conditions disappeared almost immediately once the period of exposure was stopped. As proposed by Hofmann and colleagues (1998), this likely indicates that the new audio-spatial map resulting from exposure does not interfere with the usual representation involved in normal listening. Rather, both maps are simultaneously represented in the auditory system. In concordance, two periods of exposure (i.e., two dives) were generally separated by one day or more for the participants in the present study and periods of exposure during waking hours alternated with short periods of normal listening in Javer and Schwarz (1995). Therefore, both maps would be potentially activated by auditory stimulation. The questions of how and when the auditory system selects one of the two representations remain unclear.

Conclusion

This study showed strong adaptation (plasticity) in water to smaller-than-normal ITDs and mimics past data in air showing similar adaptation to larger-than-normal ITDs. This likely means that the adult human hearing system is substantially malleable to altered ITDs regardless of the type of alteration and milieu.

This finding has direct implications for underwater navigation. We performed an informal experiment in which seven divers (above 50 hours exposure) performed four orientation swims (two based on visual signals, two based on low-frequency auditory signals) in random order. They had to reach a beacon placed at 492 feet from the starting point within 12 minutes, otherwise the swim was considered as missed. Although professional divers are only trained to visual orientation, 43% of the visual swims versus 21% of the auditory swims were missed.

More generally, every study of the effect of exposure on spatial hearing abilities with abnormal cues may have practical implications for a variety of populations. First, hearing-impaired listeners with asymmetrical cochlear loss can have elevated ITD detection thresholds (Kinkel, Kollmeïer, & Holube, 1991) and impoverished azimuthal localization performance (Durlach, Thompson, & Colburn, 1981). Second, recent studies revealed an age-related impoverishment in ITD processing before the middle life in normal-hearing adults (e.g., Moore, Vickers, & Mehta, 2012). A decrease in ITD processing possibly has a deleterious impact on spatial speech discrimination in noise (Neher, Lunner, Hopkins, & Moore, 2012). Third, various populations are concerned with virtual environments providing auditory information (e.g., video games and auditory alarms in aircrafts, medical units, and industrial or military work stations; for reviews, see Guillaume, 2011; Shilling & Shinn-Cunningham, 2002). In virtual environments, it is crucial to provide properly designed auditory information including localization cues, as spatial hearing helps awareness and orientation. One way to synthesize spatialized sounds in virtual environments is the so-called binaural synthesis. It consists in applying to the input sound different acoustic filters that mimic the effect of the head and body occurring in real listening conditions. The “real filter” can be characterized by measuring the acoustic signal at the entrance of the ear canal for a variety of source locations (i.e., by measuring the individual’s head-related transfer function, HRTF). A pair of HRTFs (both ears) contains all localization cues (spectral cues, ITDs, and ILDs). Because HRTFs are strongly listener dependent due to anatomical characteristics, the localization of virtual sources is drastically poorer using non-individual than using individual HRTFs (Wenzel, Arruda, Kistler, & Wightman, 1993). A recent study providing proprioceptive and vestibular feedback to users of a virtual auditory environment showed rapid and substantial adaptation to non-individualized HRTFs, but the authors used “hybrid” HRTFs providing non-individualized spectral cues and individualized ITDs (Parseihian & Katz, 2012). The systematic individualization of HRTFs is unachievable due to time, technical, and cost purposes. Thus, it is important that present and future research identify the mechanisms and situations by which spatial hearing abilities in altered conditions can improve up to normal.

Key Points