Fluent: An AI Augmented Writing Tool for People who Stutter
Project Overview
The document explores the innovative use of generative AI in education through the development of 'Fluent', an AI-augmented writing tool specifically tailored for individuals who stutter. This tool facilitates the composition of scripts by identifying challenging words and offering simpler alternatives, thereby enhancing fluency in speech. Utilizing an interactive interface and active learning, Fluent personalizes its suggestions based on user feedback, achieving an impressive accuracy rate of over 80% in recognizing words that may pose pronunciation difficulties. The primary objective of this tool is to bolster users' confidence in speaking scenarios, emphasizing that while it provides a helpful coping mechanism through word substitution, it does not serve as a definitive solution for stuttering. Overall, the findings highlight the potential of generative AI to create supportive educational tools that address specific communication challenges, ultimately fostering improved self-efficacy among individuals with speech difficulties.
Key Applications
Fluent - an AI-augmented writing tool
Context: Designed for individuals who stutter (PWS), especially useful in professional and personal speaking contexts like presentations and talks.
Implementation: The tool highlights difficult words in real-time and suggests easier alternatives based on user interaction and feedback, employing active learning to improve its suggestions over time.
Outcomes: Achieved over 80% accuracy in identifying difficult words with continuous improvement based on user feedback, enhancing user confidence and performance in speaking situations.
Challenges: The tool relies on user self-awareness and may not work well for individuals with acute stuttering or those requiring multiple substitutions in a sentence, potentially breaking semantic structure.
Implementation Barriers
Technical barrier
The need for users to provide seed words at the beginning requires a certain level of self-awareness, which not all users may possess.
Proposed Solutions: Future versions could include more guided support for users to help them identify their challenges.
Design limitation
The tool analyzes words individually without considering context, which may not accurately reflect pronunciation difficulties in different sentence structures.
Proposed Solutions: Future iterations could incorporate analysis of word n-grams to account for contextual pronunciation difficulties.
User engagement barrier
The effectiveness of the tool depends on the user's willingness to provide feedback, which could vary significantly among users.
Proposed Solutions: Incorporating more intuitive feedback mechanisms or prompts may encourage user participation.
Project Team
Bhavya Ghai
Researcher
Klaus Mueller
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Bhavya Ghai, Klaus Mueller
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai