Towards Natural Speech: Detection and Fluency Reconstruction of Stuttering Speech Using AI

Overview

Stuttering is a common yet challenging speech disorder affecting approximately 80 million working adults worldwide and about 3 million individuals in the United States. For many, it is a chronic condition that is difficult to fully resolve, often persisting throughout life and significantly impacting communication, psychological well-being, and educational and career opportunities. My father frequently suffers from bouts of stuttering, so much so that it impedes his speech and even his job opportunities. Witnessing his lifelong struggles inspired this research to explore how cutting-edge technologies could provide practical solutions for individuals facing similar challenges and ultimately improve their communication and quality of life. This study focused on two clinically relevant tasks: (1) accurately detecting stuttering and locating its position in speech, and (2) directly reconstructing the fluency of the dysfluent speech to improve daily communication and support speech therapy. To achieve these goals, this study developed a novel artificial intelligence (AI)-based system to analyze speech audio and perform both detection and reconstruction. For detection, the system identifies stuttering events and their temporal positions within speech. Experimental results show that the AI-based approach significantly outperforms traditional methods, achieving accuracy close to practical usability. For fluency reconstruction, a novel reference-free method was developed to restore speech by modifying dysfluent segments without requiring external training data or matched fluent recordings. Using the detected regions, the system directly reconstructs more natural-sounding speech from the original audio while preserving the speaker’s characteristics. Although no standardized benchmark currently exists for this task, evaluation on hundreds of audio clips through manual comparison indicates that the majority of reconstructed speech demonstrates clear improvement in speech fluency. This work represents a step toward practical tools for individuals who stutter, with potential applications in daily communication and speech therapy. By addressing both detection and fluency improvement, this study introduces a new direction for managing stuttering and contributes to the development of effective strategies for enhancing speech function.

Competition history

• CSEF 2026 Medicine & Physiology (Track 2) (Senior Division) · Entry S-21-30

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project