Deep multiple instance learning for foreground speech localization in ambient audio from wearable devices

Over the recent years, machine learning techniques have been employed to produce state-of-the-art results in several audio related tasks. The success of these approaches has been largely due to access to large amounts of open-source datasets and enhancement of computational resources. However, a shortcoming of these methods is that they often fail to generalize well to tasks from real life scenarios, due to domain mismatch. One such task is foreground speech detection from wearable audio devices. Several interfering factors such as dynamically varying environmental conditions, including background speakers, TV, or radio audio, render foreground speech detection to be a challenging task. Moreover, obtaining precise moment-to-moment annotations of audio streams for analysis and model training is also time-consuming and costly. In this work, we use multiple instance learning (MIL) to facilitate development of such models using annotations available at a lower time-resolution (coarsely labeled). We show how MIL can be applied to localize foreground speech in coarsely labeled audio and show both bag-level and instance-level results. We also study different pooling methods and how they can be adapted to densely distributed events as observed in our application. Finally, we show improvements using speech activity detection embeddings as features for foreground detection.

Equilibria in Personality States: A Conceptual Primer for Dynamics in Personality States

We provide a basic, step–by–step introduction to the core concepts and mathematical fundamentals of dynamic systems modelling through applying the Change as Outcome model, a simple dynamical systems model, to personality state data. This model characterizes changes in personality states with respect to equilibrium points, estimating attractors and their strength in time series data. Using data from the Personality and Interpersonal Roles study, we find that mean state is highly correlated with attractor position but weakly correlated with attractor strength, suggesting strength provides added information not captured by summaries of the distribution. We then discuss how taking a dynamic systems approach to personality states also entails a theoretical shift. Instead of emphasizing partitioning trait and state variance, dynamic systems analyses of personality states emphasize characterizing patterns generated by mutual, ongoing interactions. Change as Outcome modelling also allows for estimating nuanced effects of personality development after significant life changes, separating effects on characteristic states after the significant change and how strongly she or he is drawn towards those states (an aspect of resiliency). Estimating this model demonstrates core dynamics principles and provides quantitative grounding for measures of ‘repulsive’ personality states and ‘ambivert’ personality structures. © 2020 European Association of Personality Psychology

‘Personality in its Natural Habitat’ Revisited: A Pooled, Multi–sample Examination of the Relationships between the Big Five Personality Traits and Daily Behaviour and Language Use

Past research using the Electronically Activated Recorder (EAR), an observational ambulatory assessment method for the real–world measurement of daily behaviour, has identified several behavioural manifestations of the Big Five domains in a small college sample ( N = 96). With the use of a larger and more diverse sample of pooled data from N = 462 participants from a total of four community samples who wore the EAR from 2 to 6 days, the primary purpose of the present study was to obtain more precise and generalizable effect estimates of the Big Five–behaviour relationships and to re–examine the degree to which these relationships are gender specific. In an extension of the original article, the secondary purpose of the present study was to examine if the Big Five–behaviour relationships differed across two facets of each Big Five domain. Overall, while several of the behavioural manifestations of the Big Five were generally consistent with the trait definitions (replicating some findings from the original article), we found little evidence of gender differences (not replicating a basic finding from the original article). Unique to the present study, the Big Five–behaviour relationships were not always comparable across the two facets of each Big Five domain. © 2020 European Association of Personality Psychology

Understanding Personality through Patterns of Daily Socializing: Applying Recurrence Quantification Analysis to Naturalistically Observed Intensive Longitudinal Social Interaction Data

Ambulatory assessment methods provide a rich approach for studying daily behaviour. Too often, however, these data are analysed in terms of averages, neglecting patterning of this behaviour over time. This paper describes recurrence quantification analysis (RQA), a non–linear time series technique for analysing dynamic systems, as a method for analysing patterns of categorical, intensive longitudinal ambulatory assessment data. We apply RQA to objectively assessed social behaviour (e.g. talking to another person) coded from the Electronically Activated Recorder. Conceptual interpretations of RQA parameters, and an analysis of Electronically Activated Recorder data in adults going through a marital separation, are provided. Using machine learning techniques to avoid model overfitting, we find that adding RQA parameters to models that include just average amount of time spent talking (a static measure) improves prediction of four Big Five personality traits: extraversion, neuroticism, conscientiousness, and openness. Our strongest results suggest that a combination of average amount of time spent talking and four RQA parameters yield an R2 = .09 for neuroticism. Neuroticism is shown to be associated with shorter periods of extended conversation (periods of at least 12 minutes), demonstrating the utility of RQA to identify new relationships between personality and patterns of daily behaviour. Materials: https://osf.io/5nkr9/ . © 2020 European Association of Personality Psychology

Effects of Therapeutic Intervention on Parentally Bereaved Children's Emotion Reactivity and Regulation 15 Years Later

The Family Bereavement Program (FBP) is a family-based intervention for parentally bereaved children and surviving caregivers. Results are reported of a randomized controlled trial, examining intervention effects on emotional reactivity and regulation of young adults who participated in the program 15 years earlier. Participants (N = 152) completed four emotion challenge tasks: reactivity to negative images, detached reappraisal while viewing negative images, positive reappraisal while viewing negative images, and reengagement with positive images. Outcomes included cardiac interbeat interval (IBI), pre-ejection period (PEP), and respiratory sinus arrhythmia (RSA) as well as self-reported emotional experience and regulation effectiveness. Direct intervention effects and effects mediated through improved parenting were estimated. Several significant effects were observed in primary analyses; however, none remained significant after correction for familywise Type I error. Parenting mediated FBP effects on IBI during negative reactivity (b = 15.04), and on RSA during positive reengagement (b = 0.35); the latter effect was accounted for by changes in breathing. Intervention condition was a direct predictor of self-reported detached reappraisal effectiveness (b = 1.00). Intervention and gender interacted in predicting self-reported negative emotion during the negative reactivity (b = 1.04) and positive reappraisal tasks (b = 1.31) such that intervention-condition men reported more negative emotions during those tasks. Although these findings should be considered preliminary given the limited power of the corrected statistical tests, they suggest long-term effects of family intervention following the death of a parent on offspring's emotional reactivity and regulation ability that should be pursued further in future research.

Emotional Congruence and Judgments of Honesty and Bias

Psychological and philosophical discussions typically understand honesty as reporting truth with propositional statements. In this model, emotions are often seen as irrelevant or a hindrance to honesty, because they can bias our reports. In relational contexts, however, emotions can provide information about deep-seated convictions. We report the results of a study (N = 827) finding that individuals whose emotional responses are congruent with their explicitly stated egalitarian positions are judged as significantly more honest and less prejudiced than those with incongruent emotional responses. This is seen in judgments of white male targets who have negative emotional responses to a black man, a gay man, and a female supervisor. These results suggest that emotional reactions provide information used when judging the honesty and bias of an individual.

Going off script: Effects of awe on memory for script-typical and -irrelevant narrative detail

People often filter their experience of new events through knowledge they already have; for example, encoding new events by relying on prototypical event "scripts" at the expense of actual details. Previous research suggests that positive affect often increases this tendency. Three studies assessed whether awe-an emotion elicited by perceived vastness, and thought to promote cognitive accommodation-has the opposite effect, reducing rather than increasing reliance on event scripts. True/false questions on details of a short story about a romantic dinner were used to determine whether awe (a) reduces the tendency to impute script-consistent but false details into memory, and/or (b) promotes memory of unexpected details. Across studies we consistently found support for the first effect; evidence for the second was less consistent. Effects were partially mediated by subjective awe, and independent of other aspects of subjective affect. Results suggest that awe reduces reliance on internal knowledge in processing new events. (PsycINFO Database Record

Cognitive vulnerability to anxiety in the stress generation process: further investigation of the interaction effect between the looming cognitive style and anxiety sensitivity

BACKGROUND AND OBJECTIVES

The goal of the present study was to replicate and extend previous research on the relationship between stress generation and two well-documented anxiety related cognitive vulnerabilities, Looming Cognitive Style (LCS) and Anxiety Sensitivity (AS). We first sought to replicate findings that LCS and AS augment each other's stress generation effect. Next, we expanded upon these findings by conducting fine grained analyses not possible in the prior study, by using the third edition of the Anxiety Sensitivity Index (Taylor et al., 2007) and examined the individual facets of AS, which includes: Mental Incapacitation (fear of mental impairment), Physical (fear of catastrophic outcomes such as death), and Social (fear of being noticed for trembling, blushing) facets.

METHODS

We followed 99 female undergraduates who were assessed twice over a six-week interval.

RESULTS

First, the results replicated a previous study and showed that LCS and AS magnified each other's impact on stress generation. Second, analyses using the individual subscales of AS indicated significant interactions between LCS and the Mental Incapacitation and Physical facets of AS but not the Social facet.

LIMITATIONS

Limitations of the present study include reliance on self-report measures and the use of a female only sample. Using such a sample is consistent with previous literature, but limits generalizability to males.

CONCLUSIONS

The present findings are consistent with the emerging view that stress generation is an active, transactional process and that anxiety-related cognitive styles (much like depressive styles) contribute to stress generation.