StamFree: A Gamified AI System for Speech Disfluency Detection and Therapy in Children

Abstract

Abstract—Speech disfluency, commonly referred to as stam-
mering, is a multifaceted communication disorder that pre-
dominantly affects children between 6 to 12, with substantial

consequences for social confidence, academic achievement, and
psychosocial development. Although traditional speech therapy
demonstrates efficacy, it frequently necessitates intensive clinical
supervision and repetitive exercises, which may be stressful and
monotonous for pediatric patients, resulting in low adherence.

This study presents StamFree, an innovative, child-focused gam-
ified therapy system utilizing advanced artificial intelligence.

In contrast to earlier systems that utilize basic signal process-
ing, StamFree employs WavLM, a state-of-the-art self-supervised

deep learning model, to analyze speech directly from raw audio

waveforms. This architecture enables robust multi-class classifica-
tion of disfluencies, accurately distinguishing between repetitions,

prolongations, and blocks. The system incorporates a novel
Stress-Based Progression Strategy, which organizes phonemes
into hierarchical tiers according to articulatory stress levels:
low, medium, and high. By integrating this progression with
an adaptive unlocking mechanism, StamFree ensures that users
achieve proficiency with lower-stress sounds prior to advancing,
thereby minimizing cognitive overload. Interactive mini-games
that reinforce breathing control and pacing further contribute
to a low-anxiety, engaging therapeutic environment, promoting
sustained practice beyond the clinical context.