Auditory Rate Perception Displays a Positive Serial Dependence

Abstract

We investigated perceived timing in auditory rate perception using a reproduction task. The study aimed to test (a) whether central tendency occurs in rate perception, as shown for interval timing, and (b) whether rate is perceived independently on each trial or shows a serial dependence, as shown for other perceptual attributes. Participants were well able to indicate perceived rate as reproduced and presented rates were linearly related with a slope that approached unity, although tapping significantly overestimated presented rates. While the slopes approached unity, they were significantly less than 1, indicating a central tendency in which reproduced rates tended towards the mean of the presented range. We tested for serial dependency by seeing if current trial rate reproductions depended on the preceding rate. In two conditions, a positive dependence was observed. A third condition in which participants withheld responses on every second trial produced a negative dependency. These results suggest separate components of serial dependence linked to stimulus and response: Withholding responses reveals a negative perceptual effect, whereas making responses adds a stronger positive effect that is postperceptual and makes the combined effect positive. Together, these data show that auditory rate perception exhibits both central tendency and serial dependence effects.

Keywords: adaptation/constancy; audition; perception; temporal processing; time perception.

Figure 1.

Group mean data from 17 participants showing reproduced temporal rate plotted against presented auditory rate, with error bars showing ±1 standard deviation. Presented rates were sampled randomly from a continuous range between 3 7 Hz and have been averaged into bins with a width of 0.5 Hz. The data from the three different conditions are remarkably consistent. First, all conditions show a significant elevation above the equality line (indicated by the dashed line), and the gain in reproduced rates is less than 1 (see Table 1).

Figure 2.

Results from the serial dependence analysis for the three conditions tested showing the influence of the previous trial’s rate on current trial reproduction. Each column shows the aggregated data pooled over all participants in that condition. Trials were divided into four bins based on the frequency presented, 3–4, 4–5, 5–6, or 6–7 Hz (shown in each row), and the data within each bin were modeled with a linear regression to test for the influence of the preceding stimulus rate. The regression lines show how reproduced rates depended on the rate presented on the preceding trial. The legend in each panel shows the slope parameter and the p-value of the slope parameter from the linear fitting. In Condition 1, there was a positive dependency on previous rate for the higher frequency bins. In Condition 2, the dependence was negative, and in Condition 3, the dependency was positive. These differences are consistent with predictions and are indicative of the processes underlying serial dependency in rate perception, as discussed in the main text.

Figure 3.

The group average results of serial dependence effect for three conditions. A–C: The serial dependence is computed as the regression coefficient using a linear regression model with reproduced rates as response, and rates presented in preceding trials (1–3 back) as predictor variables. A: The computed magnitude of the serial dependence effect for rates presented one, two, and three trials back from current trial in Condition 1. The positive serial dependency lasts two trials. B: The one-back and two-back serial dependence effect in Conditions 1 and 2. In Condition 2 where observers responded on every alternating trial, a repulsive one-back serial dependence and a positive two-back serial dependence (comparable in magnitude to that in Condition 1) are observed. C: The one-back and two-back serial dependence effect in Conditions 1 and 3. The positive serial dependence is observed for one-back trial rates, but not for two-back trial rates. No strong evidence can support an enhanced serial dependence in Condition 3 than that in Condition 1. D: The influence of current trial rates on the one-back serial dependency in Conditions 1 and 3. For a given range of presented rates (3–5 Hz or 5–7 Hz), the serial dependence is defined as the difference of reproduced rates between trials preceded by 5–7 Hz rates and trials preceded by 3–5 Hz rates. The positive serial dependence is larger in Condition 3 than in Condition 1 when lower rates were presented. The magnitude of serial dependence does not differ for both conditions when higher rates were presented. Color bars are group means with ±1 standard error bars. The * symbols indicate p < .05.