Objective:To explore the changes in acoustic features of 9-10-year-old primary school students with depressive symptoms, and based on these features and the random forest (RF) algorithm, construct a model for identifying depressive symptoms in primary school students, so as to provide an intelligent psychological health screening tool for schools and education departments.Methods:This was a case-control study.A total of 1 186 primary school students aged 9-10 from three primary schools in three regions of Jiangsu Province were selected as research subjects for psychological health screening from October 26, 2022 to February 13, 2023.Their demographic data, Depression-Anxiety-Stress Scale (DASS-21) scores, Insomnia Severity Index scores, and voice recordings were collected.Based on the DASS-21 scores, the participants were divided into a control group ( n=1 086) and a depression group ( n=100).Voice recordings were made using the neutral text " The North Wind and the Sun". openSMILE was used to extract 523 acoustic features from the pre-processed voice recordings.Group differences were assessed using independent-samples t-tests or chi-square tests.Pearson correlation analysis was conducted to examine the relationship between acoustic features and depression scores.Depressive symptoms were set as the dependent variable, and the correlated acoustic features were set as the independent variable to construct a classification model using the RF algorithm.The model performance was assessed using the receiver operating characteristic (ROC) curve, the area under the curve (AUC), precision, accuracy, recall, and F1 score. Results:Compared with the control group, the depression group showed significant differences in 105 acoustic features (44 spectral features, 49 source features, and 12 prosodic features) (all P<0.05).Correlation analysis showed that 12 acoustic features (7 spectral features, 4 source features, and 1 prosodic feature) were significantly correlated with the depression score (all P<0.05).Among the RF algorithm-based classification models, the spectral features demonstrated superior performance compared to source features and prosodic features (AUC=0.793), and the performance of the model based on the combination of these features was the best (AUC=0.818). Conclusions:Acoustic features may be an objective indicator to identify the depression of 9-10-year-old primary school students, and the classification model established based on acoustic features can identify the depressed primary school students.