Toward User-Driven Sound Recognizer Personalization with People Who Are d/Deaf or Hard of Hearing
summary
Study with 14 DHH participants collects 243 recordings (avg 2.8 samples per sound) using Rev Recorder and a demo workflow built on Google’s Teachable Machine with waveform/spectrogram views. Participants find recording “easy” and engaging but face concrete problems: visualizations can mislead, playback is unreliable for those with little residual hearing, hidden sounds (e.g., sirens) are hard to capture, and users struggle to create representative, varied samples or judge model decision boundaries.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:13Prior implementations
- 0:43Enabling personalization
- 1:10Research questions
- 2:08Field Study
- 2:34Findings
- 3:32Breakdowns
- 4:31Uncertainty
- 4:59Decision Boundaries
- 5:20Summary
related talks
23mDesigning for Accessibility
20mExploring the hidden potential of sound data
12mDirection-of-Voice (DoV) Estimation for Intuitive Speech Interaction with Smart Device Ecosystems
47mThe Digital Hearth: An In-the-Wild Automated Group Speech to Text System
01h 14mExploring the hidden potential of sound data
10m#BuildPropulsionLab at #DevoxxUK: @jbaruch with @antetmc2
18mMethods for Evaluation of Imperfect Captioning Tools by Deaf or Hard-of-Hearing Users at ...
59mAccessibility for the Deaf / Hard of Hearing and Progress Over Perfection