An evidence-based toolset to capture, measure, analyze & assess emotional health

Abstract

This thesis describes the development and validation of an evidence-based toolkit that captures a patient’s emotional state, expressiveness/affect, self-awareness, and empathy during a fifteen second telephone call, and then accurately measures and analyzes these indicators of Emotional Health based on emotion detection in speech and multilevel regression analysis. An emotion corpus of eight thousand three hundred and seventy-six (8,376) momentary emotional states was collected from one hundred and thirteen (113) participants including three groups: Opioid Addicts undergoing Suboxone® treatment, the General Population, and members of Alcohol Anonymous. Each collected emotional state includes an emotional recording in response to “How are you feeling?” a self-assessment of emotional state, and an assessment of an emotionally-charged recording. Each recording is labeled with the emotional truth. A method for unsupervised emotional truth corpus labeling through automatic audio chunking and unsupervised automatic emotional truth labeling is proposed and experimented. In order to monitor and analyze the emotional health of a patient, algorithms are developed to accurately measure the emotional state of a patient in their natural environment. Real-time emotion detection in speech provides instantaneous classification of the emotional truth of a speech recording. A pseudo real-time method improves emotional truth accuracy as more data becomes available. A new measure of emotional truth accuracy, the certainty score, is introduced. Measures of self-awareness, empathy, and expressiveness are derived from the collected emotional state. Are there differences in emotional truth, self-assessment, self-awareness, and empathy across groups? Does gender have an effect? Does language have an effect? Does length of the response, as an indication of emotional expressiveness, vary with emotion or group? Does confidence of the emotional label, as an indication of affect, vary with emotion or group? Are there differences in call completion rates? Which group would be more likely to continue in data collections? Significant results to these questions will provide evidence that capturing and measuring Emotional Health in speech can: Assist therapists and patients in Cognitive Behavioural Therapy to become aware of symptoms and make it easier to change thoughts and behaviours; Provide evidence of psychotropic medication and psychotherapy effectiveness in mental health and substance abuse treatment programs; Accelerate the interview process during monthly assessments by physicians, psychiatrists, and therapists by providing empirical insight into emotional health of patients in their natural environment. Trigger crisis intervention on conditions including the detection of isolation from unanswered calls, or consecutive days of negative emotions.

Date	3 Apr 2014
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Pierre Dumouchel (Supervisor)