Automated Diagnosis and Adaptive Correction of Speech Disorders in Uzbek-Speaking Preschool Children
Ochiq kirish: yoʻq
Annotatsiya
Preschool-age children with general speech underdevelopment (GSD) represent a growing clinical population requiring individualized, consistent rehabilitation. However, access to qualified speech-language therapy remains severely limited, particularly in low-resource and non-English-speaking settings such as Uzbekistan. This paper presents an automated diagnosis and adaptive correction system for speech disorders in Uzbek-speaking preschool children that integrates three AI-driven components: (1) a Mamdani fuzzy inference engine for interpretable GSD severity classification using four multimodal input variables; (2) a Dynamic Time Warping (DTW)-based pronunciation scoring module validated against expert ratings at Pearson $\mathbf{r}=\mathbf{0. 9 1}$; and (3) a T5/mT5-based Natural Language Understanding (NLU) module fine-tuned on 10,000+ Uzbek question-answer pairs for adaptive exercise delivery. The system was deployed as a web-based clinical platform and evaluated in a single-arm pre-post pilot study (N = 21, ages 4-6, GSD levels I-III, four months) at the Republican Specialized Scientific and Practical Medical Center for Pediatrics, Tashkent. Results show a 2.0× improvement in progress rate (1.2% vs. 0.6%/session), 2.1× increase in absolute accuracy gain (58% vs. 28%), and 4.5-5.0× acceleration in speech-unit acquisition compared to the clinical baseline. These findings support the integration of NLP- and fuzzy-logic-based tools into early speech correction programs where specialist availability is limited.
Hali tarjima qilinmagan