Browsing by Author "Mehl, Matthias R."

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    Deep multiple instance learning for foreground speech localization in ambient audio from wearable devices
    (Springer, 2021) Hebbar, Rajat; Papadopoulos, Pavlos; Reyes, Ramon; Danvers, Alexander F.; Polsinelli, Angelina J.; Moseley, Suzanne A.; Sbarra, David A.; Mehl, Matthias R.; Narayanan, Shrikanth; Neurology, School of Medicine
    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.
  • Thumbnail Image
    Item
    Eavesdropping on Autobiographical Memory: A Naturalistic Observation Study of Older Adults’ Memory Sharing in Daily Conversations
    (Frontiers, 2020-06-26) Wank, Aubrey A.; Mehl, Matthias R.; Andrews-Hanna, Jessica R.; Polsinelli, Angelina J.; Moseley, Suzanne; Glisky, Elizabeth L.; Grilli, Matthew D.; Neurology, School of Medicine
    The retrieval of autobiographical memories is an integral part of everyday social interactions. Prior laboratory research has revealed that older age is associated with a reduction in the retrieval of autobiographical episodic memories, and the ability to elaborate these memories with episodic details. However, how age-related reductions in episodic specificity unfold in everyday social contexts remains largely unknown. Also, constraints of the laboratory-based approach have limited our understanding of how autobiographical semantic memory is linked to older age. To address these gaps in knowledge, we used a smartphone application known as the Electronically Activated Recorder, or “EAR,” to unobtrusively capture real-world conversations over 4 days. In a sample of 102 cognitively normal older adults, we extracted instances where memories and future thoughts were shared by the participants, and we scored the shared episodic memories and future thoughts for their make-up of episodic and semantic detail. We found that older age was associated with a reduction in real-world sharing of autobiographical episodic and semantic memories. We also found that older age was linked to less episodically and semantically detailed descriptions of autobiographical episodic memories. Frequency and level of detail of shared future thoughts yielded weaker relationships with age, which may be related to the low frequency of future thoughts in general. Similar to laboratory research, there was no correlation between autobiographical episodic detail sharing and a standard episodic memory test. However, in contrast to laboratory studies, episodic detail production while sharing autobiographical episodic memories was weakly related to episodic detail production while describing future events, unrelated to working memory, and not different between men and women. Overall, our findings provide novel evidence of how older age relates to episodic specificity when autobiographical memories are assessed unobtrusively and objectively “in the wild.”
  • Thumbnail Image
    Item
    Predicting Working Memory in Healthy Older Adults Using Real-Life Language and Social Context Information: A Machine Learning Approach
    (JMIR, 2022-03-08) Ferrario, Andrea; Luo, Minxia; Polsinelli, Angelina J.; Moseley, Suzanne A.; Mehl, Matthias R.; Yordanova, Kristina; Martin, Mike; Demiray, Burcu; Neurology, School of Medicine
    Background: Language use and social interactions have demonstrated a close relationship with cognitive measures. It is important to improve the understanding of language use and behavioral indicators from social context to study the early prediction of cognitive decline among healthy populations of older adults. Objective: This study aimed at predicting an important cognitive ability, working memory, of 98 healthy older adults participating in a 4-day-long naturalistic observation study. We used linguistic measures, part-of-speech (POS) tags, and social context information extracted from 7450 real-life audio recordings of their everyday conversations. Methods: The methods in this study comprise (1) the generation of linguistic measures, representing idea density, vocabulary richness, and grammatical complexity, as well as POS tags with natural language processing (NLP) from the transcripts of real-life conversations and (2) the training of machine learning models to predict working memory using linguistic measures, POS tags, and social context information. We measured working memory using (1) the Keep Track test, (2) the Consonant Updating test, and (3) a composite score based on the Keep Track and Consonant Updating tests. We trained machine learning models using random forest, extreme gradient boosting, and light gradient boosting machine algorithms, implementing repeated cross-validation with different numbers of folds and repeats and recursive feature elimination to avoid overfitting. Results: For all three prediction routines, models comprising linguistic measures, POS tags, and social context information improved the baseline performance on the validation folds. The best model for the Keep Track prediction routine comprised linguistic measures, POS tags, and social context variables. The best models for prediction of the Consonant Updating score and the composite working memory score comprised POS tags only. Conclusions: The results suggest that machine learning and NLP may support the prediction of working memory using, in particular, linguistic measures and social context information extracted from the everyday conversations of healthy older adults. Our findings may support the design of an early warning system to be used in longitudinal studies that collects cognitive ability scores and records real-life conversations unobtrusively. This system may support the timely detection of early cognitive decline. In particular, the use of a privacy-sensitive passive monitoring technology would allow for the design of a program of interventions to enable strategies and treatments to decrease or avoid early cognitive decline.
  • Thumbnail Image
    Item
    The paucity of morality in everyday talk
    (Springer Nature, 2023-04-12) Atari, Mohammad; Mehl, Matthias R.; Graham, Jesse; Doris, John M.; Schwarz, Norbert; Davani, Aida Mostafazadeh; Omrani, Ali; Kennedy, Brendan; Gonzalez, Elaine; Jafarzadeh, Nikki; Hussain, Alyzeh; Mirinjian, Arineh; Madden, Annabelle; Bhatia, Rhea; Burch, Alexander; Harlan, Allison; Sbarra, David A.; Raison, Charles L.; Moseley, Suzanne A.; Polsinelli, Angelina J.; Dehghani, Morteza; Neurology, School of Medicine
    Given its centrality in scholarly and popular discourse, morality should be expected to figure prominently in everyday talk. We test this expectation by examining the frequency of moral content in three contexts, using three methods: (a) Participants’ subjective frequency estimates (N = 581); (b) Human content analysis of unobtrusively recorded in-person interactions (N = 542 participants; n = 50,961 observations); and (c) Computational content analysis of Facebook posts (N = 3822 participants; n = 111,886 observations). In their self-reports, participants estimated that 21.5% of their interactions touched on morality (Study 1), but objectively, only 4.7% of recorded conversational samples (Study 2) and 2.2% of Facebook posts (Study 3) contained moral content. Collectively, these findings suggest that morality may be far less prominent in everyday life than scholarly and popular discourse, and laypeople, presume.