A Tool for Research on Emotion in Naturally Occurring Speech

Converspeech, Llc, Palo Alto CA

Investigators

Abstract

? DESCRIPTION (provided by applicant): A Tool for Research on Emotion in Naturally Occurring Speech. It is now widely accepted that defects or limitations in emotion processing underlie many disorders such as anxiety, depression, bipolar and eating disorders. If bio-behavioral markers of emotions could be found, researchers would have the basis for investigating emotions in both normal mental functioning and in a wide range of mental disorders. The role of emotion in the experience of patients with cancer and other illnesses is also of growing interest to researchers. The aim of this Phase I STTR project is to show the feasibility of developing a tool for research on emotion in naturally occurring speech. It builds o work already done by the applicant small business (Converspeech) and the applicant research organization (Stanford University). It will integrate two existing prototype programs: one that recognizes emotions in naturally occurring speech and another that analyzes language in terms of the emotions of both speaker and listener. All analyses are designed to respect individual differences in emotion processing. High-quality video and audio recordings collected from individuals and couples in interaction over two years are available to the proposed project. The project has two Specific Aims: Specific Aim 1. Build a software tool that permits acoustic and language analysis to be integrated for the study of emotion in naturally occurring speech. Specific Aim 2. Embed integrated acoustic and language analysis of speech and emotion in a tool that offers flexible choices to the researcher for new kinds of acoustic and language analysis. Feasibility will be established by using the integrated prototype tool to reproduce our recent acoustic and language study conducted on six participants (with comparable results) for four to ten additional participants. In addition, at least one new acoustic feature, one other classification model, and additional language analysis will be incorporated and tested for functionality. New language analysis will be evaluated using 50 new annotations provided by our two emotion consultants, both of whom conduct research in emotion. They will also perform a walk-through of the integrated extended prototype following a think-aloud protocol.