Exploring the challenge and strategy of identifying automatic speech recognition (ASR) errors
While speech input has improved dramatically in the past few years, reviewing and editing the dictated text during non-visual use has been a challenge for people with visual impairments. This project characterizes the challenge of identifying ASR errors in non-visual context.
Exploring the challenge and strategy of reducing teachable object recognition (TOR) errors
This project aims to build a teachable object recognizer (TOR) that helps people with visual impairments to identify objects independently using their smartphones. With a TOR, users can personalize the object recognizer by training the models with their own images. This project addresses two issues to develop this system: non-experts’ perception of machine teaching, blind users’ challenges of taking photos for training an object recognizer.