Examples using fully online / browser-based platforms

Adaptive Staircase Procedures Online

Question of Interest:

Pitch discrimination thresholds with and without a delay between tones.

Platform / Infrastructure:

Amazon Mechanical Turk. Sounds were hosted on a university server.

Participant Recruiting:

Participants in the USA and Canada.

Activities:

Participants were consented and then filled out a short demographic survey. Before they were able to complete the experiment they had to pass a brief screening to ensure that they were wearing headphones (described in Headphone screening to facilitate web-based auditory experiments. K J P Woods, M Siegel, J Traer and J H McDermott. Attention Perception and Psychophysics, vol.79 pp. 2064-2072, Jul 2017). This screening helps standardize online sound presentation.

In the main experiment, participants heard two tones (sometimes separated by a brief pause) and were asked whether the second tone was higher or lower in pitch than the first (two-alternative forced choice). We developed custom JavaScript code to perform an adaptive 1-up-X-down staircase to estimate pitch discrimination thresholds. To circumvent difficulties with rapid online sound synthesis, we pre-generated every possible pitch difference that could be heard in a given adaptive experiment. For pitch discrimination threshold measurements, we intended to increase or decrease the pitch difference in successive trials by multiplying or dividing (respectively) the initial pitch difference by 2 or sqrt(2). Thus we generated 20 possible stimuli (with different starting f0s, for example) from 8 (or more) semitones down to an undetectable pitch difference between tones, dividing by sqrt(2). For each trial, the participant would hear one of the 20 pre-generated trials at the necessary difficulty level. If participants reached the largest pre-generated pitch difference (for example, 8 semitones), the adaptive track either stopped and they were removed from the study, or participants continued to hear trials with that same pitch difference until they got a sufficient number of trials correct in a row.

Conclusions / Impressions:

We have found that pitch discrimination thresholds measured this way online are directly comparable to thresholds measured in the lab, provided basic filtering steps are taken. For example, we have sometimes used the first of several rounds of threshold estimates to remove poorly performing participants, or included a task that is orthogonal to the hypothesis being tested in order to filter poorly performing participants. A drawback of this method can be the number of files that need to be pre-generated and stored. For a standard adaptive staircase procedure this is often over 1500 audio files for a single experiment condition (20 x number of possible steps in the procedure). However, an advantage is that this method can be easily adapted to any type of adaptive threshold measurement.

Contact:

Malinda J. McPherson, mjmcp@mit.edu

Additional information and example: Efficient codes for memory determine pitch representations M J McPherson, J H McDermott. bioRxiv https://doi.org/10.1101/2020.05.07.082511 2020