StressSense

Abstract

Stress can have long term adverse effects on individuals' physical and mental well-being. Changes in the speech production process is one of many physiological changes that happen during stress. Microphones, embedded in mobile phones and carried ubiquitously by people, provide the opportunity to continuously and non-invasively monitor stress in real-life situations. We propose StressSense for unobtrusively recognizing stress from human voice using smartphones. We investigate methods for adapting a one-size-fitsall stress model to individual speakers and scenarios. We demonstrate that the StressSense classifier can robustly identify stress across multiple individuals in diverse acoustic environments: using model adaptation StressSense achieves 81% and 76% accuracy for indoor and outdoor environments, respectively. We show that StressSense can be implemented on commodity Android phones and run in real-time. To the best of our knowledge, StressSense represents the first system to consider voice based stress detection and model adaptation in diverse real-life conversational situations using smartphones.

Automatic Stress Detection in Working Environments From Smartphones’ Accelerometer Data: A First Step

“…Recently, there has also been research to detect stress outside lab environments by using wearable sensors. Lu et al [22] implemented an application running in a smartphone to detect stress using voice as input. Sano & Picard [17] used data collected from a wrist sensor, surveys and a mobile phone to classify stressed and not stressed states achieving results of over 75% accuracy.…”

“…Grouping the ratings into a smaller number of ordinal points alleviates some of the inherent subjectivity. Lu et al [22] In-lab, unconstrained, uncontrolled audio 14 subjects. accuracy of 81% and 76% for indoor and outdoor environments with model adaptation.…”

“…In recent related works it has been common to build userspecific and general models to classify stress levels [27], [22]. For the user-specific case, individual models are trained and evaluated for each of the users using their own data.…”

“…The latter approach is sometimes referred to as leave one person out. In Lu et al [22] they also used an hybrid approach called model adaptation which starts with a general model and gets adapted to each individual as more data is available.…”

New frontiers of quantified self

“…at CHI 2010-2013, BIBM 2014, UbiComp 2014), which gathered a large number of papers related to personal data, we aim to tackle some of these issues, going beyond the Quantified Self to explore both new technologies and design techniques that could be applied to this field. The UbiComp and ISWC communities are progressively designing and developing new wearable and ubiquitous systems for self-tracking and monitoring physiological parameters, such as respiratory and heart beat rate [2], behaviors, such as sleep [9], psychological states, such as stress [6]. However, the advances in technologies for automatically collecting personal information are often not matched by the critical work needed to make the gathered data more manageable, understandable, and useful for common users who are not able to manage a great quantity and variety of information.…”