The experiment was conducted following the British Psychological Society (BPS) code of practice and according to the World Medical Organization Declaration of Helsinki. All procedures had been approved by the Durham University Department of Psychology ethics committee (REF 14/13). Participants volunteered to take part in the study. Information and consent forms were provided in an accessible format, and we obtained informed consent from all participants. The current study had been conducted in the same series of work and with the same participants and facilities as one of our previous reports (Thaler et al. 2018). As such, where appropriate, we refer to method details to our previous report. Stimulus presentation and behavioural and acoustic analyses were done using Matlab (The Mathworks, Natick, USA) and custom written routines. Statistical analyses were carried out in SPSSv22.
Participants
Eight blind participants with experience in echolocation took part in the experiment. The same participants had also taken part in one of our previous studies (Thaler et al. 2018) and all details have been described in that report. For completeness, we reproduce participant details in Table 1.
Table 1 Details of participants who took part in the study Set-up and Apparatus
The work was conducted in a 2.9 m × 4.2 m × 4.9 m noise-insulated and echo-dampened room (walls and ceiling lined with foam wedges with cut-off frequency 315 Hz; floor covered with foam baffles, noise floor 24dBA). Participants stood in the centre of the room. Tactile markers were used to allow participants to reliably place their head at the same position throughout a trial, whilst not impeding movements of the mouth for clicking. Reflectors to be echolocated were wooden disks (28.5 cm or 80 cm diameter) presented one at a time in front of the participant around a reference distance of either 50 cm (28.5 cm diameter only) or 150 cm (both 28.5 cm and 80 cm diameter). Thus, at 50 cm, the 28.5-cm disk subtended 30° of acoustic angle, and at 150 cm, the 80-cm disk also comprised 30° of acoustic angle, whilst the 28.5 cm diameter at 150 cm disk comprised 10° of acoustic angle. On each trial, the reflector was first presented at the reference distance, and subsequently at a comparison distance that was either closer or farther away than the reference distance (following an adaptive staircase procedure). Reflectors were presented facing the echolocators with the centre placed at mouth level. Figure 1 illustrates the set-up. We made recordings of all testing sessions with a digital recorder (TASCAM DR100-MKII; TEAC Corporation, Japan; 24bit and 96 kHz) and with microphones (DPA SMK-SC4060 miniature microphones; 4 mm diameter; DPA microphones, Denmark) placed on either side of the participant’s head slightly in front and on top of the tragus.
General Task and Procedure
Participants placed their head in the centre of the room facing straight ahead. The head had to be kept straight ahead on the tactile marker for the whole duration of a trial. To determine distance discrimination thresholds, we employed a 2-interval-2-alternative-forced-choice adaptive staircase method. The participant’s task on every trial was to determine whether a reflector at a test distance was located closer or of a reflector at a reference distance (either 50 cm or 150 cm). Presentation was sequentially, such that the reflector was always presented first at the reference distance and then at the test distance.
To minimize the possibility of procedural bias, two intertwined staircases were used that approached each reference distance from closer or farther away (for 50 cm reference, these were − 20 or 20 cm starting value, respectively, and for the 150 cm distance, they were 60 and − 60 cm). Presentation order of staircases was pseudo-random such that one staircase would not run for more than four consecutive trials. The distance difference between test and reference on each trial was determined adaptively. In the first two trials, we used the stochastic approximation by Robbins-Monro (Robbins and Monro 1951):
$$ {x}_{n+1}={x}_n-\frac{c}{n\left({z}_n-\phi \right)} $$
where n is the number of the current trial, x the value of the stimulus (i.e. the distance of disk during test with respect to reference) and c the initial step size (set at 20 and 60 for the 50 and 150 reference distances, respectively), ϕ is the probability of responding in a correct or incorrect way with respect to the corresponding staircase (0.5 in our paradigm) and z defines if the response was correct (1) or incorrect (0), referring to the corresponding staircase (e.g. ‘closer’ is correct for the closer- and incorrect for the farther-starting staircase). For subsequent trials, we used the accelerated stochastic approximation by Kesten (1958):
$$ {x}_{n+1}={x}_n-\frac{c}{\left(2+m\right)\left({z}_n-\phi \right)} $$
which includes m for the number of changes in the response category, i.e. m increased by one when the response switched from closer to farther, or vice versa, in one staircase. The test was terminated either when the participant’s responses had reversed from closer to farther or vice versa eight times within each staircase. One session took at most 45 min to complete. There was a minimum of 15 min break in between sessions.
At the beginning of each trial, the participant would block their ears and hum. Then, the experimenter would place the reflector at the reference position. Then, the experimenter retracted to the back of the room, behind the participant. Once the experimenter was at the back of the room, they slightly tapped the foot of the participant (with a long cane). On that signal, the participant unblocked their ears and echolocated until they had a good sense of where the reflector was located (~ 6 s). Then, the participant would block their ears and hum again. The experimenter would then place the reflector at the test distance. After the reflector had been placed, the experimenter would again retract to the back of the room and signal the participant with a foot tap. The participant would then echolocate until they had a good sense of where the reflector was located (~ 6 s). Then, the participant would state whether the test was located closer or farther with respect to the reference. The experimenter would enter the response into a computer keyboard. The computer (placed in a different room) would then calculate the test position for the next trial, and display it on a monitor in the back of the testing room. During experimental trials, no feedback was given. The reason we refrained from giving feedback was that we did not want to bias participants’ behaviour in any way, and we also did not want to cause frustration. Whilst in the beginning of each staircase, i.e. for larger distance differences, perceptual differences and thus decision criteria were very clear for all participants, as the staircases progressed and distance differences became smaller (and in particular near threshold), the ‘correct’ response was expected to be less obvious and at times, all participants could do is guess.
Before the experiment started, the experimenter explained the task and procedure to the participant, and the participant completed three practice trials. During practice trials, which used the starting values for each staircase and as such were easy to perceive by all participants, the experimenter gave feedback. In addition, the participant was told that it might become increasingly more difficult to determine the position of the test with respect to the reference, and that this was a consequence of the procedure used. The participant was told that if they were uncertain about the position of the test with respect to the reference, they should respond with their ‘best guess’. The participant was also asked to keep their head stationary straight ahead placed on the tactile marker during presentation of the reflectors and in between presentation of the reference and the test. We instructed participants to give a response whenever they felt they were ready to do so (i.e. there was no limit on trial duration).
Data Analysis
Behaviour
Psychophysical performance was measured by fitting two parameter sigmoid curves of the form \( F=\frac{1}{1+\exp \left(-\frac{x-a}{b}\right)} \) to the data for each participant and test and then using these to compute thresholds and bias. Parameters were estimated by performing a non-linear least squares fit with a trust-region algorithm implemented in the Matlab optimization toolbox. Curves were fitted separately for each participant and test. To compute thresholds, we first determined those points on the curve where the probability to judge a reflector as ‘closer’ was either .25 or .75, and we then computed the average of the absolute values. To compute bias, we determined the point on the curve where the probability to judge a reflector as ‘closer’ was .5.
Acoustics
To characterize participants’ clicking behaviour, we analysed recorded sound files for each participant. We analysed the numbers of clicks made for each trial, click duration, intensity, inter-click intervals (ICIs) and click power spectra, as well as power spectral centroid, and bandwidth based on power spectra. We also computed RDLD (i.e. level difference between reflected and direct sound as measured at the ear) (Pelegrín-García and Rychtáriková 2016), echo intensity and echo power spectra. This was done to characterize participant’s echo-acoustic sensitivity. Please note that the concept of RDLD is related to the concept of target strength, as used in dolphin or bat echolocation, and which Au (1993) defined as the ratio (in dB) of the echo intensity measured 1 m from the target to the intensity of the incident signal (i.e. the emission) measured at the target. The number of clicks for each trial was determined visually and aurally by visual and aural screening of the sound files. During this process, clicks were also isolated from intermittent speech and other background noise (e.g. coughing, swallowing, etc.) for further analysis. Click duration was computed as the time from click onset to offset. To obtain onset and offset, we first computed the click envelope as the absolute value of signal and smoothing it with a moving average using a 0.42-ms-duration window. Click onset was determined as the first point where envelope value exceeded 5 % (-26 dB) of the maximum. The offset was determined as the first point where the envelope dropped to 5 % (− 26 dB) of the maximum. Click duration could only be computed for a fraction of all clicks for the condition where the reference distance was at 50 cm, because for large numbers of trials, the click duration exceeded the echo onset, meaning that the click and echo overlapped. Thus, to avoid bias in the calculation of average duration, we calculated minimum duration for clicks at all reference distances, but average click duration only for 150 cm reference distances. Click intensity was computed as root mean square (RMS) intensity of clicks for 2.6 ms from the onset of the click. Clicks were truncated at this time to avoid biasing the analysis due to overlap between click and echo. To characterize spectral content of clicks, we computed each click’s power spectrum (based on the same 2.6 ms click duration for all conditions) and then determined the power spectral centroid, and bandwidth (using a 25-dB drop relative to peak (Arditi et al. 2015), and using the powerbw.m function implemented in the Matlab signal processing toolbox) for each trial, and then averaged across trials for each distance. To compute RDLD, we determined click and echo RMS intensity, but only for those sounds where echo and click were separated in time, and then took the difference. Since reflectors had been presented straight ahead, RDLD and echo intensity were averaged across right and left channels. The echo was detected by windowing of the sound at the expected time of the echo (since the reflector had been placed at various distances), and determining RMS intensity using the same method as used for clicks. To characterize spectral content of echoes, we computed their power spectrum using the same method as used for clicks. For one participant at the 50-cm reference position, echo acoustics could not be computed because this participant’s click durations always exceeded echo onset time. For the other participants and conditions, there were sufficient ‘clean’ echo samples so that RDLD could be computed. Table 2 provides numbers of sound files used for calculation of RDLD, echo intensity and spectrum for the various conditions.
Table 2 Numbers of sounds used in each condition for calculations of RDLD and echo intensity Statistical Analysis
To investigate effects of the different conditions on thresholds and clicking behaviour, we subjected data to repeated measures ANOVA. Post hoc pairwise comparisons were done using t tests (paired samples). For all analyses, statistical significance was determined using an alpha level of .05. Greenhouse Geisser correction was applied if the sphericity assumption could not be upheld.