Digital survivor of sexual assault
summary
A deployed training system lets trainees conduct simulated conversations with a recorded survivor using a pipeline of Google speech recognition → NPC Editor classifier/dialogue manager → video player; the corpus contains 1900 video clips linked by about 19,000 example utterance-response pairs. Development uses elicitation, Wizard-of-Oz sessions at three military bases and iterative retraining over roughly a year. Evaluation with about 200 users shows the classifier usually selects appropriate clips and falls back to a four‑step non‑understanding strategy; the system is in active use at the SHARP Academy.
AI-written summary · about
chapters (6)
- 0:00Introduction to project
- 3:18Demo of survivor conversation
- 7:13How the system works
- 12:37Development process
- 15:23Deployment and results
- 16:06Q&A session
related talks
32mEmotional Trauma and Machine Learning
10mCONFIDANT: A Privacy Controller for Social Robots
20mKids in the Voice First Generation
20mSpeechGuard: Recoverable and Customizable Speech Privacy Protection...
01h 15mCHILDREN’S RIGHTS IN THE DIGITAL ENVIRONMENT: RISKS, OPPORTUNITIES, AND RESPONSIBILITIES
18mV.O.I.C.E.: A Data-Driven Risk Taxonomy for Synthetic Voice Generation
39mWhat All We've Been Up To at Knexus Research
14mTowards Understanding Privacy and Trust in Online Reporting of Sexual Assault