Abstract: Due to the lack of tone categories primarily characterized by fundamental frequency variation in Uyghur phonological system, Uyghur Putonghua learners often experience difficulties in tone acquisition, including inaccurate tone contour control, tone confusion, and insufficient neutral-tone reduction. To improve the accuracy of tone recognition for these learners, this study proposes a hybrid modeling approach that combines support vector machine (SVM) and long short-term memory (LSTM) networks. Based on a self-constructed speech corpus of Uyghur Putonghua learners, acoustic parameters such as fundamental frequency, formants, syllable duration, and differential features were extracted. SVM was first used to classify learners into pronunciation proficiency groups, and LSTM was then employed to perform five-class tone classification. Learner grouping priors were further introduced at the decision level through weighted score fusion. Experimental results show that the SVM model achieved an accuracy of 84.3% in learner proficiency grouping, while the LSTM model achieved an accuracy of 78.9% in tone classification. After incorporating learner proficiency grouping priors, the accuracy of the fusion model increased to 83.6%. These results indicate that learner proficiency priors can supplement dynamic fundamental frequency modeling and improve tone recognition performance for Uyghur Putonghua learners. This study provides technical support for automated Putonghua pronunciation assessment and instructional feedback.