Search NASA⌕ Search
Home/Archive/NASA NTRS

NASA NTRS · 20220007148

Multilevel Probit Regression for 3-Alternative Forced Choice Audibility Testing

Andrew Christian·Matthew Boucher·Menachem Rafaelof·Kevin P Shepherd·Stephen A Rizzi·James H Stephenson
View original source ↗Export metadata ↓

Abstract

A recent psychoacoustic test at NASA Langley generated a dataset of 3-alternative forced-choice responses for 40 subjects that measured the audibility of a tone complex in a shaped broadband masker. The task was completed by 4 subjects at a time in a small theatre-like environment using predetermined stimuli levels. These data were subject to 4 forms of probit regression: a “complete pooling” analysis in which all data from the test was fit with one curve, two forms of “no pooling” analyses in which subjects’ data were treated individually (using both packaged and custom software), and a “partial pooling” analysis in which multilevel-regression software fit both individual curves as well as population-level parameters at the same time. The results of the analyses are compared in terms of both individual- and population-level parameters. Partial pooling appears to give the most consistent results at both levels, as well as provide the most robustness among the packaged approaches (albeit with added complexity over single-level regression). These results are also contained in a recent NASA technical memorandum entitled “Comparisons of Analysis Methods Applied to Alternative Forced Choice Audibility Data.”

Explore related subjects

PsychoacousticsMultilevel Regression

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Andrew Christian, Matthew Boucher, Menachem Rafaelof, Kevin P Shepherd, Stephen A Rizzi, James H Stephenson. Multilevel Probit Regression for 3-Alternative Forced Choice Audibility Testing. https://ntrs.nasa.gov/citations/20220007148

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related reports

NASA NTRSDate not supplied

A Laboratory Psychoacoustic Test of the Tradeoff Between Frequent UAM Events and Their Level

Earlier this year, a psychoacoustic test was conducted in the EER which investigated how members of the public reacted to scenarios containing multiple UAM flyover events. The subjects read books of their choosing for 4 minutes at a time, during which a scenario played. They were asked to register their annoyance reaction to the scenario on an 11 point numerical scale after all of the flyovers had played. The scenarios were formulated to have several different tradeoffs between the number of flyovers played within the 4 minutes, and the level of the flyovers. The results indicate that for paces of operations typical of a busy airport (from .5 to 2 flyovers per minute), subjects seemed to be more sensitive to an increase in the number of operations than would be implied by the equal-energy hypothesis. For numbers of operations greater than this - more than what could be achieved at an airport, but perhaps what is possible in UAM scenarios - the subjects were less sensitive to increases in the number of operations and seemed to revert to the equal-energy tradeoff between level and number.

Psychoacoustics↗
NASA NTRSDate not supplied

On the Integration of Loudness Over Time for the Prediction of Single-Event Annoyance

Loudness is the quality of human perception that is most related to acoustic intensity and energy, however there is no agreed-upon way of integrating loudness over time for the prediction of noise-induced annoyance. Most noise metrics used today for aircraft certification and regulation employ the “Equal-Energy Hypothesis” (EEH) and sum the acoustical energy of one or more noise events over time to create a single metric value that represents the entire exposure. Using the EEH creates a natural tradeoff between the peak energy and the duration of a single noise event, the “hypothesis” being that this tradeoff optimally predicts annoyance in a wide variety of situations. This work proposes a strategy for integrating a loudness-like time series which is flexible in two ways: First, it includes a parameter b ∈ [0,1]. When b = 0, the metric will return the peak of the time series. At b = .5, the metric will behave in accordance with the EEH. At b = 1, the metric will penalize the duration of the sound more than the EEH would. Second, transformations are given so that the strategy can be computed from, or generate quantities in units analogous to, decibels (or phon), physical units (acoustic pressure/power), or perceptually-scaled units (sone). This approach is demonstrated on a dataset of annoyance responses to single events of UAV and road vehicle noise that is fit with an augmented linear regression. Analyses based on this integration of A-weighted level and the output of the “Zwicker” loudness model yield similar results: that subjects may have been slightly more sensitive to the durations of the events than the EEH would suppose, but that the EEH cannot be disproven using these data. The time-integrated metrics outperform both time-averaged and centile-based metrics.

Psychoacoustics↗
NASA NTRSDate not supplied

On the Integration of Loudness Over Time for the Prediction of Single-Event Annoyance

Loudness is the quality of human perception that is most related to acoustic intensity and energy, however there is no agreed-upon way of integrating loudness over time for the prediction of noise-induced annoyance. Most noise metrics used today for aircraft certification and regulation employ the “Equal-Energy Hypothesis” (EEH) and sum the acoustical energy of one or more noise events over time to create a single metric value that represents the entire exposure. Using the EEH creates a natural tradeoff between the peak energy and the duration of a single noise event, the “hypothesis” being that this tradeoff optimally predicts annoyance in a wide variety of situations. This work proposes a strategy for integrating a loudness-like time series which is flexible in two ways: First, it includes a parameter b ∈ [0,1]. When b = 0, the metric will return the peak of the time series. At b = .5, the metric will behave in accordance with the EEH. At b = 1, the metric will penalize the duration of the sound more than the EEH would. Second, transformations are given so that the strategy can be computed from, or generate quantities in units analogous to, decibels (or phon), physical units (acoustic pressure/power), or perceptually-scaled units (sone). This approach is demonstrated on a dataset of annoyance responses to single events of UAV and road vehicle noise that is fit with an augmented linear regression. Analyses based on this integration of A-weighted level and the output of the “Zwicker” loudness model yield similar results: that subjects may have been slightly more sensitive to the durations of the events than the EEH would suppose, but that the EEH cannot be disproven using these data. The time-integrated metrics outperform both time-averaged and centile-based metrics.

Psychoacoustics↗