Disfluency Generation

Disfluency generation focuses on creating artificial speech containing hesitations, repetitions, and other naturally occurring imperfections found in human conversation. Current research emphasizes leveraging large language models to generate realistic disfluent text, often employing techniques like prompt engineering and data augmentation to overcome limitations of existing, often small and imbalanced, datasets. This work is crucial for improving speech synthesis, enhancing automatic speech recognition and disfluency detection systems, and providing valuable resources for studying speech disorders like stuttering and Alzheimer's disease, where disfluencies can be diagnostic indicators.