Fig. 22.1
Spectrograms of the ‘sir’ end point of the test-word continuum showing the near sound (a) and the effects of increasing its reverberation with the 10-m room impulse response (b) which adds tails (arrowed) at the end of the vowel and at the transition from the [s] to the vowel. The effect of gating (c) is to delete the parts of these tails that extend beyond the end of the vowel
2.4 Design
The distance of the test word’s reflection pattern was 0.32 or 10 m for the ‘near’ and ‘far’ continua that were both presented in each of the four conditions. In one of the two conditions without a precursor, test words were gated to remove reverberant tails. The distance of the precursor’s reflection pattern was 0.32 or 10 m for near- and far-precursor conditions, where test words were not gated. Series of trials that each had one of the test words were presented to the six listeners in individual sessions, with 11 steps × 4 repeats × 2 test-word distances = 88 trials for each of the four conditions, all in a different randomized ordering for each listener.
The dependent variable is the difference between category boundaries for the near and far continua of a condition. This ‘near-far difference’ is reduced when a condition gives more perceptual constancy. This quantity is compared pairwise between conditions for the hypothesis tests.
2.5 Procedure
Sounds were presented to listeners at a peak level of 48 dB SPL through the left earpiece of a Sennheiser HD480 headset in the otherwise quiet conditions of the IAC booth. Before the experimental trials, listeners were informally given a few randomly selected practice trials to familiarize them with the sounds and the setup. Trials were administered to listeners in individual sessions by an Athlon 3500 PC computer with Matlab 7.1 software and with an M-Audio FireWire 410 sound card. On each of these trials, a context with an embedded test word was presented. Listeners then identified the test word with a click of the computer’s mouse, which they positioned while looking through the booth’s window at the ‘sir’ and ‘stir’ alternatives displayed on the computer’s screen. This click also initiated the following trial.
3 Results
For each condition, category boundaries for the ‘near’ and the ‘far’ continua were pooled along with the near-far difference across the 6 listeners, and the resulting means are shown with their standard errors in Fig. 22.2.
Fig. 22.2
The upper histogram shows means and standard errors of category boundaries for the near and far test words in each of the experiment’s precursor conditions. The lower histogram plots the difference between the near and far category boundaries in these precursor conditions. When this near-far difference is small, constancy is greater, so, moving from left to right, the data show constancy increasing for ‘far’ precursors, then decreasing when the precursor is removed, and then decreasing further when the test word’s tails are removed by gating in the ‘no-precursor gated’ (npg) condition
Results with near and far precursors replicate the constancy effects reported in earlier work (e.g. Watkins 2005; Nielson and Dau 2010; Watkins et al. 2011). When the precursor is nearby, increasing the test word’s distance to the far value of 10 m causes more of the continuum’s members to be heard as ‘sir’, and there is a corresponding difference between the category boundaries for near and far test words. However, when the context’s distance is also increased to 10 m, there is more constancy, as the difference between category boundaries for near and far test words is markedly reduced. The mean of the near-far differences in far-precursor conditions is significantly less than it is in near-precursor conditions, where t(5) = 14.8 and p < 0.00003. This and subsequent t-tests are two tailed with Bonferroni corrections for three comparisons.
Results in far- and no-precursor conditions allow rejection of the modulation-masking hypothesis proposed by Nielson and Dau (2010). According to this idea, any effect of removing the precursor should be a decrease in the masking of subsequent sounds, or perhaps it might have no effect at all if masking is insubstantial. The data here contradict both versions of the masking idea as the near-far difference actually increases when the far precursor is removed. This increase is considerable; the mean of the near-far differences in the no-precursor condition is significantly higher than it is in far-precursor conditions, where t(5) = 4.84 and p < 0.005. It would appear that degradation from reverberation has been restored in these no-precursor conditions, as the perceptual compensation from the precursor is removed. So the absence of degradation in Nielson and Dau’s experiment is probably due to their keeping the test word’s reverberation at the same constant level in every trial of their experiment.
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
