Audio signals generated by the human body (e.g., sighs, breathing, heart, digestion, vibration sounds) have routinely been used by clinicians as indicators to diagnose disease or assess disease progression. Until recently, such signals were usually collected through manual auscultation at scheduled visits. Research has now started to use digital technology to gather bodily sounds (e.g., from digital stethoscopes) for cardiovascular or respiratory examination, which could then be used for automatic analysis. Some initial work shows promise in detecting diagnostic signals of COVID-19 from voice and coughs. In this paper we describe our data analysis over a large-scale crowdsourced dataset of respiratory sounds collected to aid diagnosis of COVID-19. We use coughs and breathing to understand how discernible COVID-19 sounds are from those in asthma or healthy controls. Our results show that even a simple binary machine learning classifier is able to classify correctly healthy and COVID-19 sounds. We also show how we distinguish a user who tested positive for COVID-19 and has a cough from a healthy user with a cough, and users who tested positive for COVID-19 and have a cough from users with asthma and a cough. Our models achieve an AUC of above 80% across all tasks. These results are preliminary and only scratch the surface of the potential of this type of data and audio-based machine learning. This work opens the door to further investigation of how automatically analysed respiratory patterns could be used as pre-screening signals to aid COVID-19 diagnosis. CCS CONCEPTS • Information systems → Data mining; • Human-centered computing → User studies; Ubiquitous and mobile computing; • Computing methodologies → Machine learning. * Ordered alphabetically, equal contribution.

BackgroundA major cause of lapse and relapse to smoking during a quit attempt is craving triggered by cues from a smoker's immediate environment. To help smokers address these cue-induced cravings when attempting to quit, we have developed a context-aware smoking cessation app, Q Sense, which uses a smoking episode-reporting system combined with location sensing and geofencing to tailor support content and trigger support delivery in real time.ObjectiveWe sought to (1) assess smokers’ compliance with reporting their smoking in real time and identify reasons for noncompliance, (2) assess the app's accuracy in identifying user-specific high-risk locations for smoking, (3) explore the feasibility and user perspective of geofence-triggered support, and (4) identify any technological issues or privacy concerns.MethodsAn explanatory sequential mixed-methods design was used, where data collected by the app informed semistructured interviews. Participants were smokers who owned an Android mobile phone and were willing to set a quit date within one month (N=15). App data included smoking reports with context information and geolocation, end-of-day (EoD) surveys of smoking beliefs and behavior, support message ratings, and app interaction data. Interviews were undertaken and analyzed thematically (N=13). Quantitative and qualitative data were analyzed separately and findings presented sequentially.ResultsOut of 15 participants, 3 (20%) discontinued use of the app prematurely. Pre-quit date, the mean number of smoking reports received was 37.8 (SD 21.2) per participant, or 2.0 (SD 2.2) per day per participant. EoD surveys indicated that participants underreported smoking on at least 56.2% of days. Geolocation was collected in 97.0% of smoking reports with a mean accuracy of 31.6 (SD 16.8) meters. A total of 5 out of 9 (56%) eligible participants received geofence-triggered support. Interaction data indicated that 50.0% (137/274) of geofence-triggered message notifications were tapped within 30 minutes of being generated, resulting in delivery of a support message, and 78.2% (158/202) of delivered messages were rated by participants. Qualitative findings identified multiple reasons for noncompliance in reporting smoking, most notably due to environmental constraints and forgetting. Participants verified the app’s identification of their smoking locations, were largely positive about the value of geofence-triggered support, and had no privacy concerns about the data collected by the app.ConclusionsUser-initiated self-report is feasible for training a cessation app about an individual’s smoking behavior, although underreporting is likely. Geofencing was a reliable and accurate method of identifying smoking locations, and geofence-triggered support was regarded positively by participants.

The INTERSPEECH 2021 Computational Paralinguistics Challenge addresses four different problems for the first time in a research competition under well-defined conditions: In the COVID-19 Cough and COVID-19 Speech Sub-Challenges, a binary classification on COVID-19 infection has to be made based on coughing sounds and speech; in the Escalation Sub-Challenge, a three-way assessment of the level of escalation in a dialogue is featured; and in the Primates Sub-Challenge, four species vs background need to be classified. We describe the Sub-Challenges, baseline feature extraction, and classifiers based on the 'usual' COMPARE and BoAW features as well as deep unsupervised representation learning using the AUDEEP toolkit, and deep feature extraction from pre-trained CNNs using the DEEP SPECTRUM toolkit; in addition, we add deep end-to-end sequential modelling, and partially linguistic analysis.

To identify Coronavirus disease (COVID-19) cases efficiently, affordably, and at scale, recent work has shown how audio (including cough, breathing and voice) based approaches can be used for testing. However, there is a lack of exploration of how biases and methodological decisions impact these tools’ performance in practice. In this paper, we explore the realistic performance of audio-based digital testing of COVID-19. To investigate this, we collected a large crowdsourced respiratory audio dataset through a mobile app, alongside symptoms and COVID-19 test results. Within the collected dataset, we selected 5240 samples from 2478 English-speaking participants and split them into participant-independent sets for model development and validation. In addition to controlling the language, we also balanced demographics for model training to avoid potential acoustic bias. We used these audio samples to construct an audio-based COVID-19 prediction model. The unbiased model took features extracted from breathing, coughs and voice signals as predictors and yielded an AUC-ROC of 0.71 (95% CI: 0.65–0.77). We further explored several scenarios with different types of unbalanced data distributions to demonstrate how biases and participant splits affect the performance. With these different, but less appropriate, evaluation strategies, the performance could be overestimated, reaching an AUC up to 0.90 (95% CI: 0.85–0.95) in some circumstances. We found that an unrealistic experimental setting can result in misleading, sometimes over-optimistic, performance. Instead, we reported complete and reliable results on crowd-sourced data, which would allow medical professionals and policy makers to accurately assess the value of this technology and facilitate its deployment.

Online social systems are multiplex in nature as multiple links may exist between the same two users across different social media. In this work, we study the geo-social properties of multiplex links, spanning more than one social network and apply their structural and interaction features to the problem of link prediction across social networking services. Exploring the intersection of two popular online platformsTwitter and location-based social network Foursquare -we represent the two together as a composite multilayer online social network, where each platform represents a layer in the network. We find that pairs of users connected on both services, have greater neighbourhood similarity and are more similar in terms of their social and spatial properties on both platforms in comparison with pairs who are connected on just one of the social networks. Our evaluation, which aims to shed light on the implications of multiplexity for the link generation process, shows that we can successfully predict links across social networking services. In addition, we also show how combining information from multiple heterogeneous networks in a multilayer configuration can provide new insights into user interactions on online social networks, and can significantly improve link prediction systems with valuable applications to social bootstrapping and friend recommendations.