DL4DED: Deep Learning for Depressive Episode Detection on Mobile Devices

Mdhaffar, Afef; Cherif, Fedi; Kessentini, Yousri; Maâlej, M.; Thabet, J. Ben; Maâlej, M.; Jmaïel, Mohamed; Freisleben, Bernd

doi:10.1007/978-3-030-32785-9_10

Cited by 6 publications

(1 citation statement)

References 12 publications

(33 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Health assessment: Applications that analyse a patient's speech with respect to specific health conditions are meant to help the treating physicians in detecting illnesses in early stages and monitoring the disease's progress, as well as to accelerate clinical trials. Such services exist, e. g., for Parkinson's disease (Klumpp et al, 2017) and Depression (Mdhaffar et al, 2019), among others; see Cummins, Baird, and Schuller (2018); Latif, Qadir, Qayyum, Usama, and Younis (2021) for an overview on speech analysis in healthcare. Crucial in health assessment is that the resulting predictions should not replace a diagnosis by a physician, and should thus be [-automatic].…”

Section: Assessmentmentioning

confidence: 99%

Ethical Awareness in Paralinguistics: A Taxonomy of Applications

Batliner

Neumann

Burkhardt

et al. 2022

International Journal of Human–Computer Interaction

View full text Add to dashboard Cite

tations of words and phrases; it does not cover [-vocal, -verbal], i. e., facial gestures, gait, posture, and every other context, which can be attributed to the field of Affective Computing (AC), see Picard (1997). There is some overlap between CP and AC; yet, to give two examples, facial gestures are not dealt with in CP, and non-native speech as speaker characteristic is part of CP but not of AC.Arguably, most research has been done on emotion processing; early studies date back to the 1990s, e. g., on automatic speech emotion recognition (Dellaert, Polzin, & Waibel, 1996) and emotion synthesis (Schröder, 2001). Yet, this stands prototypically for other paralinguistic (and affective) phenomena such as mood, personality, and all types of typical and atypical (e. g., pathological) phenomena and variants; an overview and a sketch of the historic developments are given in Schuller and Batliner (2014). Speech Emotion Processing (SEP) encompasses generation (language), synthesis (acoustics), and recognition of emotion in speech and language. Since then, we have seen a plethora of approaches towards collecting data, modelling emotions and other paralinguistic phenomena, benchmarking with challenges, and developing applications that utilise emotional awareness. Emotion is a fuzzy term; in a prototypical use and in specific theories, it is confined to a few (four, six, or a bit more) basic emotions; in everyday use and in affect theories, less clear-cut states and traits are encompassed, e. g., interest, boredom, and frustration, and in personality theories, traits like the big five are modelled.According to some records, the "emotion-detection and -recognition market was worth $12 billion in 2018, and by one enthusiastic estimate, the industry is projected to grow to over $90 billion by 2024" (Crawford et al., 2019). In the public discourse, attention now focuses on (missing) ethical awareness. This is mirrored in the scientific community by studies that challenge basic assumptions and performance claimed, cf. Barrett, Adolphs, Marsella, Martinez, and Pollak (2019) for "inferring emotions from human facial movements". We do not know of any similar largescale study addressing the same topic in SEP or CP in general. This cannot be given in this contribution; instead, we want to introduce some ideas towards taxonomies for applications in the field; by that, we take up again and extend the taxonomy addressed in Batliner, Burkhardt, van Ballegooy, and Nöth (2006) and discuss the most important 'building bricks' of representation of states and traits and their ethical implications for real life applications.

show abstract

Section: Assessmentmentioning

confidence: 99%